MrmrXgBr: A Machine Learning Model for Breast Cancer Identification Using Next-Generation Sequencing Data
Priyadarshini A., Ratha A.K., Dash S.S., Mahanta P., Laha S.R., Nayak D.S.K.
Conference paper, Communications in Computer and Information Science, 2027, DOI Link
View abstract ⏷
The advent of artificial intelligence (AI) has brought about a significant transformation in breast cancer (BC) prediction, resulting in faster and more precise identification of cancer. Gene expression (GE) data plays a crucial role in identifying biomarkers, classifying molecular subtypes, predicting therapy responses, and uncovering therapeutic targets, all of which contribute to tailoring treatment approaches. Next-generation sequencing (NGS) outperforms microarray gene expression (GE) analysis by providing superior resolution, comprehensive assessment, and improved accuracy, thereby boosting the evaluation of risk, selection of medication, and prediction of prognosis. This research examines different machine learning (ML) methods for identifying breast cancer using NGS-GE data, with a specific emphasis on minimum redundancy maximum relevance (MRMR) for selecting features. The proposed MrmrXgBr model utilizes three widely-used ML algorithms: eXtreme gradient boosting (XGBoost), support vector machine (SVM), and random forest (RF). The selection of these algorithms was based on their strong classification performance, capability to handle data with a large number of dimensions, ability to evaluate feature importance, and capacity to identify intricate connections in GE datasets. The experimental findings showcase our proposed MrmrXgBr model's efficacy, with an impressive 96% accuracy in classification, 100% sensitivity, and 97% F1-score. These findings highlight the significance of accurate feature selection in improving the efficacy of ML models for BC detection. This research greatly enhances the application of advanced ML techniques and efficient feature engineering in creating more robust and effective tools for identifying BC. As a result, it improves patient outcomes and assists in making therapeutic decisions.
ARGai 3.0: AI fusion approach to identify antibiotic-resistant strains in E. coli using high-throughput whole genome sequencing data
Nayak D.S.K., Sucharita S., Mohanty S., Mohanty S., Bahinipati S.D., Sahoo S.K., Swarnkar T.
Article, International Journal of Data Science and Analytics, 2026, DOI Link
View abstract ⏷
The global threat of infectious diseases is being exacerbated by the issue of antimicrobial resistance (AMR). Drug resistance is growing, thus investigating antibiotic resistance strains in E. coli and urine tract infections (UTIs) is essential for effective diagnosis. Although next-generation sequencing (NGS) has made whole genome sequencing (WGS) widely available and used in AMR studies, it is still challenging to identify bacterial strain that are resistant using sequencing data. Rapid strain detection by deep learning (DL) in combating infections has reduced fatality rates. Introducing ARGai 3.0, a fusion of artificial intelligence (AI) models for resistant E. coli strain detection that is both generalized and streamlined. The ARGAi 3.0 framework for antibiotic-resistant strain classification represents a developing integration of quality control with AI models. Several classification metrics, including sensitivity, specificity, accuracy, precision, ROC-AUC, and F1-score, were assessed using ARGai 3.0 through the use of cross-validation. We hypothesized that, in comparison with baseline AI models, ARGai 3.0 would have superior capability in identifying strains of antibiotic resistance. In comparison with traditional AI models, our ARGai 3.0 demonstrates a 10% enhancement in classification accuracy and is capable of processing complex NGS WGS data. The improved sensitivity, specificity, and minimal computational time of ARGai 3.0 boost its robustness. The integration of machine learning, DL, and NGS data in ARGai 3.0 positions it as a promising tool for future antibiotic resistance surveillance. Due to its generalizability, ARGai 3.0 serves as an effective pipeline for AMR analysis.
Quantum-enhanced neural networks: bridging the quantum algorithm and machine learning
Manoswini M., Nayak D.S.K., Das T., Swarnkar T.
Book chapter, Quantum Computing, 2026, DOI Link
View abstract ⏷
Using quantum concepts like superposition and entanglement, quantum-enhanced neural networks (QENN) combine machine learning (ML) and quantum computing to improve neural network performance. They are expected to have a revolutionary influence on optimization, pattern recognition, and complex data analysis. In this work, authors discuss the principles of quantum computing, including superposition, entanglement, and quantum gates, before introducing many quantum ML (QML) techniques, including Grover’s search, Shor’s factoring, and quantum annealing. Following that, it discusses the fundamentals of neural networks for problems like image recognition and natural language processing while addressing issues with big data and high-dimensional features. This work focuses on using quantum processors to speed up traditional ML tasks by combining quantum algorithms with neural networks to generate hybrid models. It investigates quantum kernel techniques, quantum variational algorithms, and QENN, demonstrating their potential to improve conventional neural network frameworks. In addition to this authors also include real-world applications in fields including drug development, materials research, finance, and optimization to show the revolutionary effects of quantum-enhanced machine learning (QEML) on current problems. In conclusion, this chapter provides a thorough review of QML, highlighting the synergy between quantum algorithms and neural networks. It offers unique insights into the field’s current state, practical applications, and hopes for defining the future of artificial intelligence through quantum computational paradigms.
CadCNN 1.0: A HYBRID DEEP NETWORK MODEL FOR CARDIAC ARRHYTHMIA DETECTION
Patnaik B., Sahoo S., Nayak D.S.K., Subudhi A.K.
Article, Biomedical Engineering - Applications, Basis and Communications, 2026, DOI Link
View abstract ⏷
Cardiac Arrhythmia, which encompasses irregular heart rhythms such as tachycardia, bradycardia, ectopic pulses, represents a significant cause of cardiovascular morbidity and mortality worldwide, caused by disruptions in the heart’s electrical conduction system. Accurate and automated interpretation of the Electrocardiogram (ECG) is consequently crucial for the timely diagnosis and selection of treatment. This study introduces CadCNN 1.0, a novel hybrid Deep Learning (DL) framework that combines one-dimensional Convolutional Neural Networks (1D-CNN) for spatial feature extraction with Bidirectional Long Short-Term Memory (BiLSTM) for temporal sequence modeling. This methodology enables the accurate characterization of ECG waveforms and the model was trained and validated using the MIT-BIH Arrhythmia Database, classifying ECG signals into five categories: Normal (N), Fusion (F), Supraventricular Ectopic Beat (SVEB), Ventricular Ectopic Beat (VEB), and Unknown (Q) Beat. The Gray Wolf Optimization (GWO) algorithm was employed to optimize feature selection and model convergence, resulting in exceptional diagnostic performance with an Accuracy of 99.92%, Sensitivity (Recall) of 99.81%, Specificity of 99.93%, Precision of 99.92%, F1-score of 99.92% along with MCC of 99.84% with Number of Wolves (Nw) of 10 represents the number of wolves in the population, corresponding to the number of candidate feature subsets simultaneously explored in each iteration, that enhances population diversity and prevents premature convergence and Max Iteration of 10 which denotes the maximum number of optimization cycles during which the wolves update their positions toward the optimal feature subset, ensuring iterative refinement, stability of global optimum. The confusion matrix demonstrates CadCNN 1.0’s discriminative ability to differentiate morphologically equivalent Arrhythmic classes with near-perfect precision, as evidenced by its strong diagonal dominance and minimal off-diagonal misclassifications. Furthermore, the Breast Cancer Dataset performed cross-domain validation, resulting in an Accuracy of 97.50%, Sensitivity (Recall) of 96.49%, Specificity of 97.18%, Precision of 96.49%, F1-score of 96.49%, MCC of 92.53%. These results underscore the model’s adaptability to heterogeneous biomedical data. CadCNN 1.0 provides a clinically reliable, generalizable, computationally efficient computer-aided diagnostic system that advances automated Arrhythmia detection and enables intelligent, patient-centered healthcare diagnostics by integrating spatial-temporal feature learning, meta-heuristic optimization, and interpretable confusion-matrix validation.
Blockchain Technology: Foundations, Challenges, and Future Prospects
Meher S.K., Bahinipati S.D., Mohanty B., Nayak D.S.K.
Book chapter, Decentralized Futures: Navigating the Integration of Blockchain Technology with Artificial Intelligence, 2026, DOI Link
View abstract ⏷
Blockchain technology appears to be the solution that many companies seek in a safe, decentralized, and transparent way. Originally hailed as the building block of the Bitcoin blockchain, its effect is no longer just limited to the cryptocurrency industry but has branched out to the healthcare, governmental, supply chain, and banking sectors. It is a detailed exploration of the blockchain technology – its architecture, basic concepts, and consensus techniques. The decentralized, immutable, transparent, and secure validation features of this system provide a trusted infrastructure for safe online transactions. An analysis of actual deployments of a variety of blockchain frameworks: public, private, consortium, and hybrid. This chapter looks at some consensus approaches like Practical Byzantine Fault Tolerance (PBFT), Delegated Proof of Stake (DPoS), Proof of Work (PoW), Proof of Stake (PoS), and some other more current alternatives along with PoW, PoS, and PBFT. The chapter looks into the pros and cons of blockchain concerning scalability, energy consumption, and regulation, followed by an analysis of recent advances to overcome these issues. This expert study should aid in future research and industry implementation.
Deep Network Models for Alzheimer’s Diagnosis Using MRI: A Comparative Analysis of CNN and Hybrid CNN-LSTM Architectures
Panda S., Singh V., Dash S.S., Ratha A.K., Meher S., Nayak D.S.K.
Conference paper, Lecture Notes in Networks and Systems, 2026, DOI Link
View abstract ⏷
Alzheimer’s disease (AD) is a degenerative brain condition that impairs cognitive function, erodes memory, and alters behavior. The key to managing this disease lies in early detection. This study introduces a cutting-edge deep learning framework for detecting AD using Magnetic resonance imaging (MRI) data. In this paper, we have taken MRI images of both dementia and non-dementia subjects. We presented two architectures: a Convolutional Neural Network (CNN) and a Hybrid model. The hybrid model combines CNN and LSTM layers to analyze sequential dependencies in the data. We have done a comparison analysis between these two models. Our results show that the hybrid CNN-LSTM model outperforms the CNN model, achieving a classification accuracy of 98.70% compared to 98% for the CNN model. Besides, we have done a comparative study of ROC-AUC curve and confusion matrix to show the execution of the model. Again, at training procedure, model loss and accuracy also depicted according to epoch times. The researches already performed on prediction of Alzheimer’s disease has given poor performance of 94% classification accuracy in comparison to our model.
ViTat 1.0: An attention based vision transformer model for brain tumor identification
Swain R.P., Bahinipati S.D., Kharya S., Das T., Patnaik B., Nayak D.S.K.
Book chapter, Futuristic Information and Communication: A Multimodal Multidisciplinary Signal Analysis, 2026, DOI Link
View abstract ⏷
Deep learning algorithms are essential for rapid and reliable diagnosis of brain tumors (BT), a crucial task in medical imaging. Here, we offer a comparison method to investigate how well Vision Transformers and Convolutional-Neural-Networks (ViCNNs) perform in improving the detection of brain tumors. Our research shows encouraging outcomes, suggesting substantial improvements in model performance and classification accuracy. We proposed a DL model ViTat 1.0 enriched with attention-based Vision Transformer (ViT) mechanism to detect medical images’ global dependencies. In addition to this, we also tested ViTat 1.0 and other studied CNN models on Generative Adverbial Networks (GANs) augmented images and observed a significantly increase (more than 5%) in classification accuracy. The results show Consistent and higher performance, which confirms that the proposed ViTat 1.0 is practical across different datasets. Results show that ViTat 1.0 exceeds CNNs, proving once again how critical it is to use cutting-edge DL architectures for medical image processing. Finally, our study enhances the field of BT identification by demonstrating the effectiveness of GAN and attention with ViT in improving classification accuracy and model performance. With these results, ViTat 1.0 can be seen as an application that could improve medical image processing and lead to faster, more precise clinical diagnosis.
DeepCCi 1.0: A Deep Learning Model for Cervical Cancer Identification
Mahanta P., Mohanty S., Priyadarshini A., Ratha A.K., Dash S.S., Nayak D.S.K.
Conference paper, Lecture Notes in Networks and Systems, 2026, DOI Link
View abstract ⏷
Cervical cancer (CC) arises from abnormal cell proliferation in the cervix, a vital anatomical part of the uterus. It is caused by the human papillomavirus (HPV). CC is the largest cause of mortality among women in underdeveloped nations, emphasizing the critical need for early identification and care. To reduce the burden of CC, it is critical to guarantee early detection and treatment under expert medical supervision. Traditional diagnostic approaches frequently struggle with unbalanced datasets, resulting in poor model performance. This research examines several Artificial Intelligence (AI) models for the identification of CC with a balanced dataset, which is obtained by adopting the Synthetic Minority Over-sampling Technique (SMOTE). In addition to this, we also test the efficacy of Artificial Intelligence (AI) on selected features on the balance dataset. For generalization purposes, we validate our proposed DeepCCi 1.0 model by taking different datasets. We also benchmark our proposed DeepCCi 1.0 model with existing state-of-the-art models. Our proposed DeepCCi 1.0 model overcomes the limitations, especially classification accuracy and computational time, as compared to several traditional approaches. The experimental findings showcase our proposed DeepCCi 1.0 model’s efficacy, with an impressive classification sensitivity, accuracy, F1-score, specificity of 99%, and MCC of 98%. These outcomes highlight the significance of accurate feature selection in improving the efficacy of ML models for CC identification. This research greatly enhances the application of ML and DL techniques and efficient feature engineering in creating more robust and effective tools for identifying CC. As a result, it improves patient outcomes and assists in making therapeutic decisions.
YouTube Sentiment Analyzer for Modi (YSAM): Unlocking YouTube Comment Sentiments with Machine Learning and Visualization
Panda S., Dash S.S., Ratha A.K., Sahoo S., Sasmal A., Nayak D.S.K.
Conference paper, International Conference on Innovations in Intelligent Systems: Advancements in Computing, Communication, and Cybersecurity, ISAC3 2025, 2025, DOI Link
View abstract ⏷
Sentiment analysis has emerged as a critical field for understanding public opinion through textual data analysis. This study presents a comprehensive pipeline for sentiment analysis of YouTube comments regarding Indian Prime Minister Narendra Modi, utilizing machine learning techniques to identify complex patterns in public sentiment. The proposed methodology provides an effective approach for understanding citizen opinions through social media discourse analysis. The pipeline encompasses data extraction, preprocessing, vectorization using Term Frequency-Inverse Document Frequency (TF-IDF) and Bag of Words techniques, followed by classification using multiple machine learning models. A dataset comprising 3,000 comments was divided into training, validation, and testing subsets, with over-sampling techniques employed to address class imbalance issues. The ExtraTrees Classifier demonstrated superior performance, achieving 96% accuracy in sentiment classification. Additionally, this study presents a user-friendly sentiment analysis dashboard for real-time processing and visualization of sentiment patterns, highlighting the potential of machine learning applications in sentiment analysis and providing valuable insights for researchers and practitioners in computational social science.
aiGeneR 3.0: an enhanced deep network model for resistant strain identification and multi-drug resistance prediction in Escherichia coli causing urinary tract infection using next-generation sequencing data
Nayak D.S.K., Pati A., Panigrahi A., Khan M., Alabdullah B., Sahoo S.K., Sahu B., Almjally A., Mallik S., Swarnkar T.
Article, Frontiers in Genetics, 2025, DOI Link
View abstract ⏷
Background: Infectious diseases pose a global health threat, with antimicrobial resistance (AMR) exacerbating the issue. Considering Escherichia coli (E. coli) is frequently linked to urinary tract infections, researching antibiotic resistance genes in this context is essential for identifying and combating the growing problem of drug resistance. Objective: Machine learning (ML), particularly deep learning (DL), has proven effective in rapidly detecting strains for infection prevention and reducing mortality rates. We proposed aiGeneR 3.0, a simplified and effective DL model employing a long-short-term memory mechanism for identifying multi-drug resistant and resistant strains in E. coli. The aiGeneR 3.0 paradigm for identifying and classifying antibiotic resistance is a tandem link of quality control incorporated with DL models. Cross-validation was adopted to measure the ROC-AUC, F1-score, accuracy, precision, sensitivity, specificity, and overall classification performance of aiGeneR 3.0. We hypothesized that the aiGeneR 3.0 would be more effective than other baseline DL models for antibiotic resistance detection with an effective computational cost. We assess how well our model can be memorized and generalized. Results: Our aiGeneR 3.0 can handle imbalances and small datasets, offering higher classification accuracy (93%) with a simple model architecture. The multi-drug resistance prediction ability of aiGeneR 3.0 has a prediction accuracy of 98%. aiGeneR 3.0 uses deep networks (LSTM) with next-generation sequencing (NGS) data, making it suitable for novel antibiotics and growing resistance identification in the future. Conclusion: This work uniquely integrates SNP-level insights with DL, offering potential clinical utility in guiding antibiotic stewardship. It also enables a robust, generalized, and memorized model for future use in AMR analysis.
Artificial intelligence in improving disease diagnosis: A case study of cardiovascular disease prediction
Pati A., Addula S.R., Panigrahi A., Sahu B., Kumar Nayak D.S., Dash M.
Book chapter, Artificial Intelligence in Medicine and Healthcare, 2025, DOI Link
View abstract ⏷
Every year, cardiovascular disease (CVD) accounts for one-third of the world's fatalities. Premature death affects 43% of the poor, compared to 7% of the rich. The cause is lifestyle-related disorders such as diabetes and obesity. There was a low rate of premature mortality, which highlights the need for early cardiac illness identification. It is crucial to combine biochemical and clinical data for the early detection of cardiac disease. By giving cardiologists access to state-of-the-art data analysis and therapeutic decision-making tools, artificial intelligence (AI) treatments can potentially revolutionize the field. AI tools, such as deep learning (DL) and machine learning (ML), can aid medical professionals in learning more complicated and massive data sets. Early detection of heart disease is very necessary. In this paper, we have developed two-stage experiments on an integrated five-heart disease dataset from the UCI-ML warehouse. First, we tested 13 ML and DL algorithms on the integrated dataset. In the second stage, we applied these conventional classifiers to the featured dataset obtained by applying sequential feature selection and observed the enhanced predictive outcomes with the same integrated dataset as 97.06% accuracy, 98.81% precision, 97.08% sensitivity, 97.01% specificity, and 97.94% F1 score that justifies the novelty of the proposed model. Finally, ROC curves with AUC values were added to the study's empirical analysis, and an improved AUC of 0.987 was recorded. This proposed work might help physicians diagnose heart diseases.
DeepEcc 1.0: Ensemble Deep Learning Model Enriched with Whale Optimization for Cervical Cancer Identification
Mahanta P., Mishra A., Taram M., Sahu S., Jaswal A., Nayak D.S.K.
Conference paper, 2025 IEEE 6th Global Conference for Advancement in Technology, GCAT 2025, 2025, DOI Link
View abstract ⏷
The fourth most common malignancy worldwide is cervical carcinoma (CC). HPV causes CC. In developing countries, HPV kills women. HIV-positive women account for roughly 5% of cervical cancer cases. So early detection is crucial. Traditional HPV testing and colposcopy treatment outcomes are superior. This study uses AI models to detect cervical cancer early, as they are most powerful, fast, and cost-effective. This study uses whale optimization and 1-DCNN/Bi-LSTM ensemble models to detect cervical cancer. Traditional identification processes struggle with feature optimization. Whale optimization using an ensemble deep learning model to recognize CC was our powerful weapon. By employing whale optimization, finding the best solutions and improving efficiency. The work uses VGG16, VGG19, ResNet50, and ResNet101 AI models. These popular deep learning architectures can extract important information and capture subtle patterns, improving classification accuracy. The combination of those two approaches outperforms CNN or LSTM alone, enhances accuracy, and is resilient. Experimental results demonstrate that the proposed model can handle raw, imputed, and whale-optimized gene expression data. DeepEcc 1.0 outperforms CNN designs like VGG-16, VGG-19, ResNet-50, and ResNet-101 in accuracy, sensitivity, specificity, and MCC. DeepEcc 1.0 handles high-dimensional and imbalanced datasets well with an accuracy of 85% and an MCC of 76, especially on the whale-optimized dataset. Benchmarking against recent studies using TCGA and GEO datasets shows that DeepEcc 1.0 is on par with or better than existing state-of-the-art approaches for identifying cervical cancer based on gene expression data. DeepEcc 1.0 can detect cervical cancer early and accurately thanks to Whale optimization and ensemble DL architecture, which improve feature selection and model generalization.
HPVC 1.0: Computational Intelligence in Human Papillomavirus Classification – A Machine Learning and Deep Learning Paradigm
Swain S., Parida M., Mohanty R.N., Swain J., Kumar Nayak D.S., Meher S.
Conference paper, International Conference on Artificial Intelligence and Emerging Technologies, ICAIET 2025, 2025, DOI Link
View abstract ⏷
HPV type categorization is essential for early detection and determining the risk of developing HPV-related cancers. The novelty of the proposed work is an intelligent system, HPVC 1.0, that uses ML and DL approaches to classify HPV types in different classes by exploiting gene size features obtained from viral genomes of various strains of HPV genotypes. SMOTE and ADASYN address the class imbalance issue in the dataset. Altogether, we evaluated eight models in the following categories: machine learning (ML) algorithms such as Support Vector Machine (SVM), Logistic Regression, Random Forest, K-Nearest Neighbour(KNN), Decision Tree and deep learning (DL) models including Artificial Neural Networks(ANN), Convolutional Neural Networks(CNN), Long Short-Term Memory(LSTM) and Gated Recurrent Unit(RU). Performance HPVC 1.0 is expressed in accuracy, F1-score, sensitivity, specificity, and ROC-AUC. Results show that SVM and Logistic Regression are more stable classifiers on small datasets, whereas the Random Forest model and CNN possess higher accuracy in larger balanced datasets. This HPVC 1.0 system provides a scalable and efficient approach to the classical HPV type identification, with the potential applications for real-time clinical diagnosis and public health screening programs
Diabetes Prediction in Healthcare Utilizing Machine Learning Algorithms
Tripathy N., Hota S., Satapathy P., Nayak S.K., Nayak D.S.K., Nayak P.P.
Conference paper, 2025 International Conference on Ambient Intelligence in Health Care, ICAIHC 2025, 2025, DOI Link
View abstract ⏷
Diabetes mellitus is among the most dangerous conditions that many people suffer from. Age, obesity, poor food, heredity, high blood pressure, and inactivity are some of the variables that might lead to diabetes mellitus. Though there are additional forms of diabetes, such as gestational diabetes, which appears during pregnancy, the two most common types are type 1 and type 2. In data science, machine learning is a relatively new field of study that looks at how machines pick up knowledge via experience. The goal of this work is to build a system that can more accurately forecast a patient's risk of developing diabetes in their early years by utilizing a combination of research data and machine learning techniques. Interpretable findings using logistic regression (LR) show how various risk variables affect the chance of acquiring diabetes. The most crucial characteristics for diabetes prediction may be automatically discovered by neural networks (NN) from unprocessed input data, eliminating the requirement for human feature selection. While compared to previous models, both the logistic regression and the neural network in this study provide a respectable accuracy score following a 10-fold classification.
BCViMer 1.0: A Deep Network Pipeline for the Identification of Breast Cancer Using MRI Data
Das T., Ratha A.K., Dash S.S., Pandey R., Mohanty S., Nayak D.S.K.
Conference paper, 3rd IEEE International Conference on Industrial Electronics: Developments and Applications, ICIDeA 2025, 2025, DOI Link
View abstract ⏷
Breast cancer ranks among the foremost causes of cancer-related deaths in women globally, with early diagnosis via advanced imaging techniques significantly enhancing patient outcomes. Convolutional neural networks (CNNs) demonstrate considerable potential in the automation of breast cancer detection. Nonetheless, their constrained capacity to capture global image features may hinder their performance. This study presents BCViMer 1.0, a deep learning pipeline utilizing the Vision Transformer (ViT) to overcome existing limitations and enhance breast cancer classification accuracy from MRI images. This research aimed to assess the efficacy of BCViMer 1.0 relative to established CNN architectures, specifically ResNet-50, ResNet-101, VGG-16, and VGG-19. Experimental results indicate that BCViMer 1.0 excels these CNN models, attaining an average classification accuracy that is 4.15% higher. The enhanced performance of BCViMer 1.0 is due to its capacity to capture long-range dependencies and global context in MRI images, facilitated by the self-attention mechanism of the ViT. In conclusion, BCViMer 1.0 exhibits superior diagnostic accuracy compared to conventional CNN-based models, representing a significant advancement in breast cancer detection. This study emphasizes the capabilities of Vision Transformers in medical imaging and establishes a basis for future investigation in clinical applications.
A Prioritized-Recommendation System with Association Rule Mining and Random Forest for Retailing
Nayak D.S.K., Nanda P., Agasti P., Anjum R., Harichandan A.K., Laha S.R.
Conference paper, Proceedings of the 2025 International Conference on Artificial Intelligence and Emerging Technology, Global AI Summit 2025, 2025, DOI Link
View abstract ⏷
This work proposes a Machine Learning-based Prioritized-Recommendation System for the retail industry. Traditional recommendation systems often face challenges in ranking items effectively, which can lead to lower engagement and sales. This research addresses this gap by integrating Association Rule Mining (ARM) with a Random Forest classifier to predict the likelihood of customer purchases based on transaction data. The system first extracts item relationships using ARM and creates feature vectors from these rules and customer data. A Random Forest model is then used to predict purchase probabilities, and prioritization ranks recommendations accordingly. The proposed system archives model accuracy of 94.96%, Precision 0.85, Recall 0.893, F1-Score 0.88, and AUC-ROC 0.93, demonstrating superior performance in recommending the most relevant products. Furthermore, the use of Incremental ARM ensures the model can dynamically adapt the changes in customer behavior, providing an efficient and scalable solution for personalized retail marketing.
Lung Cancer Identification with Deep Networks: Convolutional Neural Network or Transformers!
Behera S.R., Dash S.S., Dash A.A., Mishra P.P., Anjum R., Nayak D.S.K.
Conference paper, ICRTEECT 2025 - 2nd International Conference on Recent Trends in Electrical, Electronics and Computing Technologies, 2025, DOI Link
View abstract ⏷
Lung cancer is one of the most common and deadly cancers worldwide, with low survival rates due to delayed identification. Timely and accurate diagnosis can improve treatment outcomes and reduce mortality. Histopathological imaging remains the gold standard for diagnosis, but manual inspection is laborious and subjective. In this context, deep learning approaches can automate classification and support pathologists in making reliable decisions. This research presents a deep learning-based multi-class lung cancer classification framework using histopathological images. The dataset consists of three categories: benign, adenocarcinoma, and squamous cell carcinoma. Multiple CNN architectures, including MobileNetV2, ResNet-101, and EfficientNet-B3, were evaluated. Among them, MobileNetV2 achieved the highest classification accuracy of 99.1%, while EfficientNet-B3 achieved 97.2% with a strong balance across precision, recall, F1-score, and ROC-AUC. To address class imbalance and overfitting, transfer learning, class weighting, and data augmentation were applied. Confusion matrix analysis further confirmed the ability of the models to distinguish between similar cancer subtypes. The findings highlight the potential of EfficientNet-based CNNs for reliable medical image analysis and digital pathology. Future work will explore Vision Transformers and attention mechanisms to enhance diagnostic accuracy and interpretability. Overall, this research demonstrates the promise of AI-supported histological evaluation and its role in advancing clinical decision support.
GenENet 1.0: A Deep Network Pipeline with Generative Adversarial Networks for Tumor Identification
Das T., Mohanty S., Nayak D.S.K., Ratha A.K., Swain S., Swarnkar T.
Conference paper, International Conference on Innovations in Intelligent Systems: Advancements in Computing, Communication, and Cybersecurity, ISAC3 2025, 2025, DOI Link
View abstract ⏷
Brain tumors must be diagnosed early and accurately in order to be treated effectively and with timely intervention. Despite the significant potential of deep learning in medical image analysis, its performance can be impeded by factors such as class imbalance and lack of data availability. In response to these challenges, we introduce GenENet 1.0, a deep learning pipeline that mixes InceptionV3 with GANs to enhance data quality and enable strong tumor classification. There is a significant reduction in data imbalance and an improvement in model generalization when GANs are utilized to generate a variety of synthetic samples. InceptionV3, renowned for its exceptional feature extraction, enhances the accuracy of categorization. GenENet 1.0 showed remarkable sensitivity and specificity of 97.50% when evaluated on a benchmark dataset of brain tumors. In addition, our model outperforms the state-of-the-art methods according to ROC analysis, which yielded an AUC-ROC score of 97.35%. The combination of GAN-based data augmentation and the modified InceptionV3 architecture enhances tumor detection, lowers misclassification, and increases diagnostic reliability. These results demonstrate that GenENet 1.0 is a powerful platform for identifying brain cancers and illustrate that automated medical image processing has come a long way. We endeavour to expand the dataset, refine the model, and integrate multi-modal medical imaging approaches in order to enhance real-world clinical application in the future.
Feline Emotion Recognition Using Deep Learning: A Comprehensive EfficientNet-B0 Framework for Multi-Class Sentiment Analysis in Domestic Cats
Bhattacharya A., Verma R.K., Nayak D.S.K., Panda S., Sasmal A., Sahoo S.
Conference paper, 2025 1st International Conference on Advancement in Futuristic Technologies, ICAFT 2025, 2025, DOI Link
View abstract ⏷
Recognizing emotional states in domestic cats presents significant challenges due to the subtle nature of feline facial expressions and behavioral cues. This research introduces a robust deep learning framework for automated feline emotion recognition across seven distinct emotional categories: angry, disgusted, happy, normal, sad, scared, and surprised. Leveraging the EfficientNet-B0 architecture with transfer learning, our model achieves exceptional classification performance with 99.20% training accuracy and 99.67% validation accuracy on a comprehensive dataset of feline facial expressions. The methodology incorporates strategic data partitioning, advanced augmentation techniques, and weighted random sampling to address class imbalance. Additionally, comprehensive statistical validation through parametric and non-parametric tests confirms the robustness and generalizability of our approach. Experimental results demonstrate near-perfect classification metrics across all emotional categories, with precision, recall, and F1-scores of 0.99 or higher. This research advances the field of animal-computer interaction by providing veterinarians, animal behaviorists, and pet owners with an accurate, automated tool for understanding feline emotional states, thereby contributing to improved animal welfare and enhanced human-feline relationships.
Optimized Deep Learning Architecture for Maize Disease Detection Using Efficient Channel Attention and Transfer Learning
Dash L., Laha S.R., Pattanayak B.K., Nayak D.S.K., Chakraborty P., Sarkar S.
Conference paper, 2025 Global Conference on Information Technology and Communication Networks, GITCON 2025, 2025, DOI Link
View abstract ⏷
Maize is a major staple food crop worldwide. However, its yield is threatened by foliar diseases, such as northern corn leaf blight, common rust, and gray leaf spot, due to which timely disease detection is essential to ensure food safety and minimize pesticide usage. We propose a hybrid deep-learning model by combining a CNN trunk (7 & 7 convolution, batch normalization, ReLU) with EfficientNet inverted-bottleneck blocks, along with Efficient Channel Attention (ECA) to improve cross-channel feature calibration in a lightweight model with a competitive approach for 4.5M parameters. PlantVillage maize recall images were resized to 224&224 (Maize-leaf-PV) and augmentation (rotations, flips, zooms, brightness, and channel shifts) to mimic field diversity. The model was trained using the Adam optimizer with a cosine-annealing schedule for 350 epochs on TPU pods, batch size 64, dropout rate 0.5, early stopping, and model checkpointing to avoid overfitting. The model achieved an accuracy of 98.85 with F1-scores of 93.37 (Blight), 99.45 (Common Rust), 89.57 (Gray Leaf Spot), and 99.90 (Healthy). These are significantly better than state-of-the-art baselines, including ResNet-50 (95.58%), Inception-v3 (95.99%), and EfficientNetB2 (94.00%), while decreasing the computational cost and model memory footprint. The lightweight structure is conducive to real-time application in a precision-agriculture system and allows targeted response to reduce crop yield losses and environmental harm. Towards implementation: robustness validation under diverse field conditions with UAV and smartphone imaging, transfer learning for additional crops, ensemble and severity-estimation modules, optimization for mobile-device implementation to democratize access to automated disease monitoring, collaborative farmer feedback loops, and cloud-based deployment pipelines for continuous model improvements.
Facial Emotion Recognition for University Students using CNN: Transforming Learning Environment
Swain A., Laha S.R., Sahoo S., Dalei A., Srivastav V., Kumar Nayak D.S.
Conference paper, ICoICC 2025 - 3rd International Conference on Intelligent and Cloud Computing, 2025, DOI Link
View abstract ⏷
The well-being and academic performance of university students are significantly influenced by their emotional experiences within learning environments. This research presents a novel application of Facial Emotion Recognition (FER) using web cameras and Convolutional Neural Networks (CNN) to enhance classroom and laboratory settings. The system is designed to monitor students' emotional states in real time, enabling educators to adapt learning environments for improved engagement and well-being. Initially, an image dataset was collected through web cameras placed in classrooms, capturing students' facial expressions during lectures and activities. The dataset was preprocessed to account for variations in lighting, facial orientations, and occlusions. A CNN-based model was developed and trained to classify emotions such as happiness, sadness, anger, and neutrality. Hyperparameter tuning and data augmentation techniques were employed to optimize model performance. The model was evaluated using metrics such as accuracy, precision, recall, and F1-score to ensure its effectiveness across diverse conditions. Additionally, the system was optimized to handle real-time emotion detection in varying classroom environments. This real-time FER system allows continuous monitoring of students' emotional states, providing actionable insights for educators to create more supportive and adaptive learning environments. The results demonstrate the potential of FER technologies to transform educational environments by fostering a deeper understanding of student emotions and promoting well-being. This research offers a foundation for the integration of AI-driven tools in modern education to enhance learning experiences.
Deep Networks and Internet of Medical Things for Tracking the Post Surgical Recovery Condition: A Comparative Approach
Samal S.M., Sibadutta Mishra A., Bose M., Swain A., Laha S.R., Kumar Nayak D.S.
Conference paper, 2025 2nd IEEE International Conference for Women in Computing, InCoWoCo 2025, 2025, DOI Link
View abstract ⏷
Proper post-operative recovery tracking is crucial for the early identification of complications and optimal patient outcomes. This paper introduces a computer-vision-based, sensor less technique fusing DL models with the Internet of Medical Things (IoMT) to predict a patient's recovery status Healthy or Unhealthy. Estimated vital signs such as heart rate (HR), blood pressure (BP), temperature, SpO2, and hemoglobin are obtained from facial photographs and video inputs by applying machine learning algorithms. A synthetic dataset of 10,000 features is reduced in dimension to less than 2,000 dimensions by using Principal Component Analysis (PCA) and autoencoders. Three models are compared: a Baseline Deep Neural Network (DNN), Autoencoder + Multi-Layer Perceptron (MLP), and Residual MLP. Among these, Residual MLP yields the best classification accuracy of 94.7%, beating others because of improved gradient flow in deep architectures. A Streamlit-enabled web interface makes real-time vitals estimation and prediction possible. The findings highlight the viability of contactless, AI-based post-operative monitoring and stress the need for both feature compression and sound model selection. Real-world dataset expansion and mobile health deployment remain future directions.
Machine Learning-Driven Classification Framework Using ViT and Swin Transformers for Feature Extraction of Brain MRI Images
Panigrahi S., Adhikary D.R.D., Jena P.P., Biswal U., Dash A., Nayak D.S.K.
Conference paper, International Conference on Artificial Intelligence and Emerging Technologies, ICAIET 2025, 2025, DOI Link
View abstract ⏷
Brain tumors are life-threatening conditions characterized by the uncontrolled proliferation of abnormal cells. Accurate and timely detection is essential, as delayed diagnosis may result in severe outcomes. Magnetic Resonance Imaging (MRI) remains the most widely used modality for brain examination due to its superior soft tissue contrast and resolution. Over the last decade, research in brain MRI analysis has evolved from conventional machine learning approaches to advanced deep learning frameworks. In this work, we examine the potential of Vision Transformer (ViT) and Swin Transformer architectures for brain tumor classification. The ViT model applies transformer mechanisms to image patches, enabling global contextual learning without convolutional operations. In contrast, the Swin Transformer introduces a hierarchical representation with shiftedwindow self-attention, allowing scalable and locality-aware feature extraction. We propose a hybrid strategy that integrates features from both pre-trained ViT and Swin models and evaluates them with eight classical machine learning classifiers. Experiments were conducted on the publicly available Br35H: Brain Tumor Detection 2020 dataset to assess classification accuracy and computational performance under limited data conditions. The results are validated using multiple evaluation metrics to provide a comprehensive performance analysis.
Attention enhanced hybrid deep learning model with 1D-CNN and BiLSTM for automated sleep apnea detection
Article, Discover Applied Sciences, 2025, DOI Link
View abstract ⏷
Sleep Apnea (SA) is an alarming sleep disorder characterized by repeated cessation of breathing during sleep, which frequently results in severe cardiovascular and metabolic impairment. The traditional techniques, such as polysomnography, are time-consuming, costly, and require clinical supervision. To address these challenges, this study proposes a hybrid DL-based framework for automated SA identification with single-lead ECG signals from the PhysioNet Apnea-ECG dataset. The methodology integrates comprehensive signal processing along with feature engineering to improve the morphological, temporal characteristics of ECG signals. The traditional ML models like SVM with Linear, Polynomial, Sigmoid, Radial Basis Kernels, and DL architectures such as 1D-CNN, 1D-CNNN-BiLSTM, and a hybrid 1D-CNN + BiLSTM-Attention mechanism automatically weight temporal features to focus on diagnostically significant ECG segments and improve the hidden feature space representation of prominent patterns. The efficacy of the model is demonstrated in the form of a validated quantitative performance evaluation using 10-fold cross-validation. The model achieved 98.39% Accuracy, 99.02% Precision, 98.29% Sensitivity, 96.53% Specificity, 98.66% F1-Score, 96.67% MCC, 97.78% AUC on PhysioNet-Apnea ECG dataset, indicating strong classification performance with robust generalization across patient’s sub-groups. The strength of the proposed model has been demonstrated using the MIT-BIH Polysomnographic database across various sleep patterns and recording conditions in diagnostic accuracy, efficiency, and generalization across physiological variations. The results have been compared with some of the state-of-the-art methods for establishing the superiority of the proposed model.
Vi-GeN 1.0: GAN-Augmented Vision Transformer Pipeline for Diversified Pneumonia Classification
Dash A., Panigrahi S., Nayak D.S.K., Dash A.P., Swarnkar T.
Article, Journal of Transformative Technologies and Sustainable Development, 2025, DOI Link
View abstract ⏷
Pneumonia continues to be one of the global causes of morbidity and mortality, necessitating early and proper diagnosis using chest X-rays (CXRs). Deep learning networks, specifically convolutional neural networks (CNNs), have shown excellent results in identifying pneumonia. But their performance is normally hampered by dataset imbalance, feature redundancy, and local receptive fields, reducing generalizability. To overcome such challenges, the study introduces Vi-GeN 1.0, a GAN-augmented Vision Transformer (ViT) pipeline specifically architected for pneumonia classification robustness. Our method uses Wasserstein GAN with Gradient Penalty (WGAN-GP) to synthesize high-quality CXRs, improving dataset diversity and feature learning. The ViT classifier, trained on the real and synthetic data, learns global contextual representations, resulting in better classification performance. Vi-GeN 1.0 model was tested on the Kaggle Chest X-ray Pneumonia dataset, showing 97.3% accuracy, 98.1% sensitivity, 96.5% specificity, and AUC-ROC of 0.985, performing much better than the baseline ViT model (p < 0.01). The realism and feature preservation of the synthetic images are ensured by FID (22.1), SSIM (0.87), and PSNR (28.9 dB) validation. Our results showed that GAN-based augmentation improves pneumonia classification performance effectively, making Vi-GeN 1.0 a valuable candidate for clinical use, especially in low-resource environments. Future research will investigate multi-class classification, domain generalization, and more sophisticated augmentation methods like diffusion models.
Boosted Query Expansion for Agricultural Decision Support: A Hybrid Framework Combining Case-Based Reasoning, Fuzzification, and Machine Learning مئاقلا للادتسلاا نيب جمدي نيجه لمع راطإ:يعارزلا رارقلا معدل تاملاعتسلال ززعُملا عيسوتلا ةللآا مّلعتو ،ةيبابضلا ةيمعتلاو ،تلااحلا ىلع
Solanki S., Srivastav V., Bhattacharya A., Roy P., Laha S.R., Kumar S., Nayak D.S.K.
Article, Tikrit Journal of Engineering Sciences, 2025, DOI Link
View abstract ⏷
This framework, “BQ-CBRS,” Hybrid Bigger Query-Case Based Reasoning System model, is the first of its kind to unite contextual embeddingbased query expansion (using BERT), IndRNN-based semantic similarity scoring, the fuzzification of uncertain parameters, and XGBoost classification within one application to support precision agriculture. Some of the steps include query preprocessing, generating contextual embeddings utilizing a pre-trained method (for example, BERT), semantic similarity scoring using IndRNN, and expanding the query by adding top-ranked search terms. Fuzzification will acknowledge any uncertainty present in the data, while XGBoost will enhance the predictive power and efficacy of the present work. The proposed methodology consists of query preprocessing, contextual representations using pre-trained models (like BERT), calculating a similarity score through IndRNN, and expanding the query according to the top-scoring terms. Fuzzification will address the uncertainty in the data, and XGBoost will enhance prediction accuracy and efficiency. The Crop Recommendation Dataset consists of parameters, such as nitrogen, phosphorus, pH, temperature, and rainfall. The present model has low accuracy and low mean square error (MSE). Also, it improves over traditional approaches. The model will utilize precision agriculture technology to link historical cases and improve approaches for more effective resource management and advancing sustainable farming. This combination of symbolic reasoning and deep learning in the agriculture domain is novel, establishing a generalizable framework for intelligent decision support in dynamic and uncertain situations.
Evaluating the Capability of Large Language Model Chatbots for Generating Plain Language Summaries in Radiology
Sarangi P.K., Hui P., Mondal H., Kumar Nayak D.S., Swarup M.S., Ishan, Panda S.
Article, iRADIOLOGY, 2025, DOI Link
View abstract ⏷
Background: Plain language summary (PLS) are essential for making scientific research accessible to a broader audience. With the increasing capabilities of large language models (LLMs), there is the potential to automate the generation of PLS from complex scientific abstracts. This study assessed the performance of six LLM chatbots: ChatGPT, Claude, Copilot, Gemini, Meta AI, and Perplexity, in generating PLS from radiology research abstracts. Methods: A total of 100 radiology abstracts were collected from PubMed. Six LLM chatbots were tasked with generating PLS for each abstract. Two expert radiologists independently evaluated the generated summaries for accuracy and readability, with their average scores being used for comparisons. Additionally, the Flesch–Kincaid (FK) grade level and Flesch reading ease score were applied to objectively assess readability. Results: Comparisons of LLM-generated PLS revealed variations in both accuracy and readability across the models. Accuracy was highest for ChatGPT (4.94 ± 0.18) followed by Claude (4.75 ± 0.31). Readability was highest for ChatGPT (4.83 ± 0.27) followed by Perplexity (4.82 ± 0.29). The Flesch reading ease score was highest for Claude (62.53 ± 10.98) and lowest for ChatGPT (40.10 ± 11.24). Conclusion: LLM chatbots show promise in the generation of PLS, but performance varies significantly between models in terms of both accuracy and readability. This study highlights the potential of LLMs to aid in science communication but underscores the need for careful model selection and human oversight.
ARGai 1.0: A GAN augmented in silico approach for identifying resistant genes and strains in E. coli using vision transformer
Nayak D.S.K., Das R., Sahoo S.K., Swarnkar T.
Article, Computational Biology and Chemistry, 2025, DOI Link
View abstract ⏷
The emergence of infectious disease and antibiotic resistance in bacteria like Escherichia coli (E. coli) shows the necessity for novel computational techniques for identifying essential genes that contribute to resistance. The task of identifying resistant strains and multi-drug patterns in E. coli is a major challenge with whole genome sequencing (WGS) and next-generation sequencing (NGS) data. To address this issue, we suggest ARGai 1.0 a deep learning architecture enhanced with generative adversarial networks (GANs). We mitigate data scarcity difficulties by augmenting limited experimental datasets with synthetic data generated by GANs. Our in-silico method (augmentation with feature selection) improves the identification of resistance genes in E. coli by using feature extraction techniques to identify valuable features from actual and GAN-generated data. Employing comprehensive validation, we exhibit the effectiveness of our ARGai 1.0 in precisely identifying the informative and resistant genes. In addition, our ARGai 1.0 identifies the resistant strains with a classification accuracy of 98.96 % on Deep Convolutional Generative Adversarial Network (DCGAN) augmented data. Additionally, ARGai 1.0 achieves more than 98 % of sensitivity and specificity. We also benchmark our ARGai 1.0 with several state-of-the-art AI models for resistant strain classification. In the fight against antibiotic resistance, ARGai 1.0 offers a promising avenue for computational genomics. With implications for research and clinical practice, this work shows the potential of deep networks with GAN augmentation as a practical and successful method for gene identification in E. coli.
PerSceptoMed 1.0: An advanced machine learning model for demographic-based antibiotic susceptibility prediction in Pseudomonas aeruginosa infections
Routray S.P., Mahapatra S., Nayak D.S.K., Sahoo S., Das B.S., Subudhi E., Panigrahy R., Swarnkar T.
Article, Engineering Applications of Artificial Intelligence, 2025, DOI Link
View abstract ⏷
Background: Antimicrobial resistance caused by Pseudomonas aeruginosa is a pressing global health issue impacting both hospitals and communities. Methodology: We analyzed demographic and susceptibility data from 1087 samples, and tested them against five different antibiotics at the Institute of Medical Sciences & SUM Hospital. Eight machine learning algorithms (linear: multiple linear regression, generalized linear model-elastic net regularization, support vector machine-linear kernel, linear discriminant analysis; non-linear: support vector machine-radial basis function kernel, decision tree, k-nearest neighbors, kernel k-nearest neighbors), were evaluated to develop the PerSceptoMed 1.0 model for susceptibility prediction. Performance metrics such as accuracy, Matthews Correlation Coefficient, Hamming loss, 0/1 loss, and Receiver Operating Characteristic curves were computed. Statistical analysis was conducted to compare model performance. Results: Gender disparities were observed, with male patients showing higher infection rates, potentially due to occupational exposure, microbiome differences, comorbidities, and healthcare-seeking behaviors. age, clinical status, and antibiotic susceptibility were significant factors, complicating treatment strategies. Infections extended to outpatient departments, indicating a shift from hospital to community settings. Kernel k-nearest neighbors exhibited strong predictive capabilities, improving accuracy from 84.39% pre-balancing to 91.01% post-balancing, and an Area Under the Curve of 0.92. Statistical tests showed significant performance variations among models. Conclusion: The study highlights the importance of gender-sensitive approaches in infection management and the superior performance of nonlinear models in decoding complex data patterns. PerSceptoMed 1.0 demonstrates promise in enhancing antimicrobial resistance management by providing actionable insights based on demographic data.
Prevalence of multidrug-resistant Klebsiella pneumoniae in urinary tract infections: A retrospective observational study in eastern India
Sahoo S., Mohanty J.N., Routray S.P., Sarangi A., Nayak D.S.K., Shah S., Das J., Subudhi E., Swarnkar T.
Article, Microbes and Infectious Diseases, 2025, DOI Link
View abstract ⏷
Background: Urinary tract infection (UTI) commonly affects people of different ages. It is important to explore the antibiotic susceptibility of the bacterial agents to improve the empirical antibacterial prescription to tackle the fast emerging of multi-drug resistant (MDR) bacteria. Klebsiella pneumoniae, a Gram-negative bacterium, is widely known for its hypervirulence, drug resistance and opportunistic nature, with significant implications for critically ill or immunocompromised patients. The study aim is to determine the prevalence of antibiotic resistance pattern among isolates from urinary tract infections (UTIs) and provide valuable insights for healthcare management. Methods: This is a retrospective observational study including 250 patients among 500 sample screened to have UTI at the largest hospital in eastern India from July 2020 to November 2022. All UTI patients underwent agar-based bacterium identification and antibiotic sensitivity testing using the disc diffusion method. Results: Klebsiella pneumoniae (122 cases) and Escherichia coli (74) were the first and second most frequent Gram-negative isolates, followed by Pseudomonas aeruginosa (45) and Acinetobacter baumannii (9). Notably, ten Klebsiella pneumoniae strains were resistant to five major antibiotic classes: β-lactam, aminoglycosides, fluoroquinolones, tetracyclines, and polymyxin. piperacillin/tazobactam, ceftazidime, aztreonam, and imipenem exhibited high resistance rates when tested against Klebsiella pneumoniae isolates. In contrast, gentamicin, levofloxacin, minocycline, fosfomycin, and colistin demonstrated relatively lower effectiveness against Klebsiella pneumonia isolates. Conclusion: Monitoring of multidrug-resistant bacteria becomes critical for prudent antibiotic use, ultimately dropping the prevalence of hospital-acquired infections and contributing to the larger fight against antimicrobial resistance.
ARGai 2.0: A Feature Engineering Enabled Deep Network Model for Antibiotic Resistance Gene and Strain Identification in E. coli
Nayak D.S.K., Priyadarshini A., Routray S.P., Sahoo S.K., Swarnkar T.
Article, International Journal of Online and Biomedical Engineering, 2025, DOI Link
View abstract ⏷
Escherichia coli (E. coli) is a group of bacteria that cause infections in the gastrointestinal (GI) tract and urinary tract (UTIs). The rise in antimicrobial resistance (AMR) due to antibiotic-resistant genes (ARGs) linked to E. coli strains that cause UTIs poses a significant threat. Identifying ARGs and resistant strains is crucial for effective treatment. To identify ARGs and classify resistant strains in E. coli utilizing gene expression (GE) data with advanced computational techniques such as feature engineering and transfer learning (TL). In TL, knowledge acquired by the baseline model is transferred to the target domain (BI-LSTM-GRU). ARGai 2.0 utilizes the Synthetic Minority Over-sampling Technique (SMOTE) for over-sampling the GE dataset to evaluate the effectiveness of the proposed TL framework. Our proposed ARGai 2.0 model achieved a higher classification accuracy of 14% compared to 1D CNN and 11% compared to BI-LSTM-GRU individually. The analysis revealed that genes associated with the nitrate reductase operon (narU, narV, narW, narY, and narZ) exhibit high connectivity and interaction scores, indicating their central role in nitrate metabolism. This aligns with the high enrichment FDR of 3.07E-10 and fold enrichment of 229.28 for pathways related to nitrate reductase complex and nitrite transmembrane transporter activity. ARGai 2.0 successfully detected the significant genes responsible for antibiotic resistance and classified the resistant strains. The gene network analysis highlights the central role of nitrate metabolism genes, while peripheral genes like ansP and yncG are involved in more specialized functions.
An Artificial Intelligent Approach for Classifying Odia and Non-Odia Handwritten Script using Histogram of Oriented Gradients (HOG) Features
Tripathy S., Pattanaik S.R., Nayak D.S.K.
Conference paper, 2nd International Conference on Cognitive, Green and Ubiquitous Computing, IC-CGU 2025, 2025, DOI Link
View abstract ⏷
Handwritten script identification plays an important role in Optical character recognition (OCR), specially when multilingual and regional language settings like Odia are present. A comparative analysis of K-Nearest Neighbours (KNN) and Support Vector Machine (SVM) classifiers for distinguishing between Odia and Non-Odia handwritten characters can be found here. Two feature representations are being analysed: raw greyscale image pixels and Histogram of Orientated Gradients (HOG) descriptors for better accuracy. Our Experimental results demonstrate that HOG features significantly enhance classification performance. The SVM classifier attains up to maximum accuracy of 91.99% utilising HOG, whereas KNN usually achieves 88.44%. Using raw image data yields lower accuracy (i.e. 55.33 for SVM and 44.20 % for KNN), on the other hand. Importance of skilful feature extraction for improving model performance is emphasised in this study. SVM exceeds KNN because of its resilience in high-dimensional environments. From this we can conclude that by using affective preprocessing on dataset, the performance of model accuracy can be significantly high and it also suggests available options, such as dataset augmentation and deep learning.
CiGeN 1.0: A Generative AI Enabled Medical Image Augmentation Pipeline for Deep Network Models
Conference paper, 2025 International Conference on Ambient Intelligence in Health Care, ICAIHC 2025, 2025, DOI Link
View abstract ⏷
Deep Learning methods have gained a significant momentum in medical image classification tasks, including many ranges of datasets. However, these datasets often suffer from class imbalance, where the minority class (e.g., diseased cases) is underrepresented. This imbalance leads to biased predictions, affecting the model's performance. Generative Adversarial Networks (GANs) offer a potential solution by generating synthetic images for the minority class, helping balance the dataset. Methodology: In this study we further evaluated our Deep Learning Model CoVaD-GAN with separate set of datasets to test the dynamics behind it. We applied a Deep Convolutional GAN (DCGAN) to generate synthetic images for underrepresented classes in the chest X-ray (CXR) and Break His histopathological dataset. Augmented synthetic images were made a part of training set, and the model's performance was evaluated before and after augmentation using VGG-16, ResNet-50 with necessary metric measures as accuracy, sensitivity, and specificity were compared pre- and post-GAN augmentation. We also evaluated the quality of GAN-generated images using Fréchet Inception Distance (FID) and manual expert reviews. The results demonstrate significant improvements in model accuracy, sensitivity, and F1-scores after applying GAN-based augmentation. The use of GANs for data augmentation enhances the ability of deep learning models to generalize, particularly for minority classes. GAN-generated images were of high quality and contributed positively to model performance. The Proposed model (CiGeN 1.0) dynamically generalizes the other medical dataset introduced to it and derived a fair resulting diagnosis.
Identification of Key Genetic Drivers in Multidrug Resistance: A Co-Expression Network Analysis of Pseudomonas aeruginosa
Routray S.P., Mahapatra S., Nayak D.S.K., Subudhi E., Swarnkar T.
Conference paper, 2025 International Conference on Ambient Intelligence in Health Care, ICAIHC 2025, 2025, DOI Link
View abstract ⏷
Antimicrobial resistance (AMR) in Pseudomonas aeruginosa, a critical pathogen in healthcare-associated infections, demands new insights into underlying genetic mechanisms. This study aimed to identify key regulatory genes contributing to multidrug resistance (MDR) through a comprehensive analysis of gene expression in clinical isolates. RNA-Seq data from 246 clinical isolates of P. aeruginosa was analyzed using Weighted Gene Co-Expression Network Analysis (WGCNA) to construct co-expression networks. Modules of co-expressed genes were identified, and hub genes with high connectivity were selected, focusing on those within and across modules, including the largest module enriched with resistance-related genes. Pathway enrichment was performed using STRING and ShinyGO, highlighting key resistance-associated pathways. Pathway enrichment analysis of hub genes identified polA as a central player in DNA repair and replication processes, crucial for bacterial survival under antibiotic stress. In addition to polA, other highly connected hub genes, such as dnaB, tatC, and ppnK, were enriched in pathways including protein transport and mismatch repair. Within the turquoise module, hub genes like dnaA, uvrD, and gyrA were primarily involved in ATP-binding, ligase activity, and biosynthetic processes, essential for antibiotic resistance. Network analysis further emphasized the centrality of polA and its role in bridging multiple resistance pathways, highlighting its potential as a therapeutic target. This research highlights polA and other critical hub genes as key regulatory elements within the MDR networks of P. aeruginosa, playing essential roles in pathways vital for bacterial survival. The identification of these genes provides promising biomarkers and potential therapeutic targets to counteract MDR in clinical settings.
aiGeneR 1.0: An Artificial Intelligence Technique for the Revelation of Informative and Antibiotic Resistant Genes in Escherichia coli
Nayak D.S.K., Mahapatra S., Routray S.P., Sahoo S., Sahoo S.K., Fouda M.M., Singh N., Isenovic E.R., Saba L., Suri J.S., Swarnkar T.
Article, Frontiers in Bioscience - Landmark, 2024, DOI Link
View abstract ⏷
Background: There are several antibiotic resistance genes (ARG) for the Escherichia coli (E. coli) bacteria that cause urinary tract infections (UTI), and it is therefore important to identify these ARG. Artificial Intelligence (AI) has been used previously in the field of gene expression data, but never adopted for the detection and classification of bacterial ARG. We hypothesize, if the data is correctly conferred, right features are selected, and Deep Learning (DL) classification models are optimized, then (i) non-linear DL models would perform better than Machine Learning (ML) models, (ii) leads to higher accuracy, (iii) can identify the hub genes, and, (iv) can identify gene pathways accurately. We have therefore designed aiGeneR, the first of its kind system that uses DL-based models to identify ARG in E. coli in gene expression data. Methodology: The aiGeneR consists of a tandem connection of quality control embedded with feature extraction and AI-based classification of ARG. We adopted a cross-validation approach to evaluate the performance of aiGeneR using accuracy, precision, recall, and F1-score. Further, we analyzed the effect of sample size ensuring generalization of models and compare against the power analysis. The aiGeneR was validated scientifically and biologically for hub genes and pathways. We benchmarked aiGeneR against two linear and two other non-linear AI models. Results: The aiGeneR identifies tetM (an ARG) and showed an accuracy of 93% with area under the curve (AUC) of 0.99 (p < 0.05). The mean accuracy of non-linear models was 22% higher compared to linear models. We scientifically and biologically validated the aiGeneR. Conclusions: aiGeneR successfully detected the E. coli genes validating our four hypotheses.
Weighted Bayesian Belief Network for diabetics: a predictive model
Kharya S., Soni S., Pati A., Panigrahi A., Giri J., Qin H., Mallik S., Nayak D.S.K., Swarnkar T.
Article, Frontiers in Artificial Intelligence, 2024, DOI Link
View abstract ⏷
Diabetes is an enduring metabolic condition identified by heightened blood sugar levels stemming from insufficient production of insulin or ineffective utilization of insulin within the body. India is commonly labeled as the “diabetes capital of the world” owing to the widespread prevalence of this condition. To the best of the authors' last knowledge updated on September 2021, approximately 77 million adults in India were reported to be affected by diabetes, reported by the International Diabetes Federation. Owing to the concealed early symptoms, numerous diabetic patients go undiagnosed, leading to delayed treatment. While Computational Intelligence approaches have been utilized to improve the prediction rate, a significant portion of these methods lacks interpretability, primarily due to their inherent black box nature. Rule extraction is frequently utilized to elucidate the opaque nature inherent in machine learning algorithms. Moreover, to resolve the black box nature, a method for extracting strong rules based on Weighted Bayesian Association Rule Mining is used so that the extracted rules to diagnose any disease such as diabetes can be very transparent and easily analyzed by the clinical experts, enhancing the interpretability. The WBBN model is constructed utilizing the UCI machine learning repository, demonstrating a performance accuracy of 95.8%.
Enhanced Seizure Prediction Using XGBoost and EEG Data: A Machine Learning Approach with increasing Model Metrics
Panda S., Dash S.S., Pandey R., Sahoo S., Sasmal A., Nayak D.S.K.
Conference paper, 2nd International Conference on Signal Processing, Communication, Power and Embedded Systems, SCOPES 2024, 2024, DOI Link
View abstract ⏷
Epilepsy is a neurological disorder marked by abnormal brain activity. Normally, brain cells (neurons) communicate with each other through electrical signals. In epilepsy, these signals become disrupted, leading to excessive and synchronized firing, resulting in seizures. But, if it can be predicted early, then it may be prevented or made to progress slowly. For prediction, machine learning has played a great role nowadays. In this study, we have used models like Random Forest, Decision Tree, Extra-trees classifier, Gradient Boost (XGBoost) has given better result for prediction. In this paper we have taken total 500 individuals EEG data with each has 4097 data points. The EEG scans provided detailed images of brain structures, allowing for precise analysis of epileptic activity using EEG data. The prediction of seizures has proven to be a formidable challenge, with existing studies failing to yield consistently accurate results. Our methodology involved a 60/40 train-test split and again 20/20 for test-validation split ensuring the model was trained on the majority of the data. By utilizing the XGBoost, we achieved an accuracy of 98%, demonstrating its effectiveness in handling complex spatial data inherent in EEG scans. Besides, we have presented roc-auc curve, f1-score, recall, support, precision and confusion matrix to show the execution of the model. This localization capability represents a significant advancement in epilepsy research, enabling more targeted treatment strategies. The researches already performed on prediction of Epilepsy disease has given poor performance of 95% classification accuracy in comparison to our model.
Modified Deep Neural Network Approach to Identify Heart Disease using IoMT: Artificial Neural Networks or Convolutional Neural Networks!
Nayak D.S.K., Priyadarshini A., Mahanta P., Panigrahi S., Meher S., Swain S.
Conference paper, 2024 International Conference on Intelligent Computing and Sustainable Innovations in Technology, IC-SIT 2024, 2024, DOI Link
View abstract ⏷
Unconventional methods of predicting and treating heart disease (HD) are required because HD is still the top mortality on a global scale. This study delves into the potential of combining deep learning (DL) methods with IoMT devices to improve HD prediction and early diagnosis. Despite their usefulness, traditional risk assessment approaches can misrepresent the intricate relationship between cardiovascular risk factors. On the other hand, DL and ML algorithms can sift through mountains of data related to lifestyle, genetics, demographics, and clinical conditions to find complex relationships. To increase generalization performance across varied patient demographics and datasets, our study is focused on designing a novel DL architecture that is specialized for HD prediction. We optimize its design and hyperparameters. With feature engineering, when selecting features, we use linear discriminant analysis (LDA). In our work, two different DL techniques neural network and convolutional neural network (CNN) for HD identification are studied. We then evaluate the studied model's efficacy with measures including recall, accuracy, precision, F1-score, and Matthews Correlation Coefficient (MCC). By attaining 96% accuracy, 97% precision, 95% specificity, 97% F1-score, and 93% MCC on chosen features, the experimental findings show that a one-dimensional (1-D) CNN performs better than other models. Our suggested method, which combines DL and IoMT, offers a powerful answer to the problems of real-time health monitoring and HD risk assessment, which in turn improves early intervention and tailors treatment plans to each patient. The findings of this study highlight the promise of cutting-edge AI-driven approaches to revolutionizing cardiovascular healthcare.
Optimizing XGBoost for Enhanced Electrical Theft Detection: A Fusion of Particle Swarm and Jaya Optimization Techniques
Prusty S., Nayak D.S.K., Moharana M., Prusty S.G.P., Sahoo M., Rautaray J.
Conference paper, 2024 IEEE 4th International Conference on Sustainable Energy and Future Electric Transportation, SEFET 2024, 2024, DOI Link
View abstract ⏷
The most serious and hazardous for an electrical provider these days are non-technical losses caused due to electricity theft. The economy as a whole is impacted by fraudulent electricity usage, which lowers supply quality, raises generator load, and drives up costs for legal customers. To avoid these issues, this work proposes a smart electrical theft detection system using the XGBoost model combined with two optimization techniques Particle Swarm Optimization (PSO) and Jaya Optimization (JO). Furthermore, to handle the missing instances in the State Grid Corporation of China (SGCC) imbalanced dataset, we implemented a novel LORAS (Local Outlier-Robust Adaptive Synthetic) technique that balances the target class by making fake data points near the edges of different classes. Finally, the results show that the optimized XGBoost-PSO technique significantly classifies into two distinct categories (i.e. theft as '1' or no theft as '0') with having higher F1 Score (PSO Optimized): 0.9552 and AUC (PSO Optimized): 0.9667.
Web-Based Tools Validation for Antimicrobial Resistance Prediction: An Empirical Comparative Analysis
Routray S.P., Sahoo S., Nayak D.S.K., Shah S., Swarnkar T.
Article, SN Computer Science, 2024, DOI Link
View abstract ⏷
Antimicrobial resistance (AMR) is a serious threat to global public health, necessitating rapid and precise diagnostic tools. The prevalence of novel antibiotic resistance genes (ARGs) has increased due to microbial sequencing, resulting in the need to extract vital information from vast amounts of data. Although many AMR prediction tools exist, only a few are accurate and scalable. We examined 20 widely used AMR prediction tools and chose 4 web-based tools for antimicrobial resistance surveillance over standalone software due to their easy accessibility, portability, and centralized data management, eliminating the need for complex installation and maintenance. CGE (Center for Genomic Epidemiology) provides bioinformatics tools and promotes open data sharing. At the same time, CARD (Comprehensive Antibiotic Resistance Database) is a valuable resource for antibiotic resistance gene information, collectively contributing to our understanding and management of antibiotic resistance. We highlighted web-based AMR prediction tools and performed a case study using the Pseudomonas aeruginosa complete plasmid sequence (CPS) to identify strengths and weaknesses in the system. Our study explored four web-based antibiotic resistance gene prediction tools: ResFinder, KmerResistance, ResFinderFG, and RGI. ResFinder excelled at finding acquired antimicrobial resistance genes as well as maintaining a database up to date. KmerResistance identified resistance genes using k-mer analysis. esFinderFG offered a unique perspective, excelling in detecting a broad range of resistant phenotypes, due to its inclusion of sequences discovered through functional metagenomics. RGI was versatile in detecting a wide range of resistance genes and provided extensive resistance mechanism information. Researchers must understand the capabilities and trade-offs of these tools to make well-informed choices for efficient resistance gene identification and surveillance as the antibiotic resistance landscape evolves.
An Investigation into Opportunities and Challenges for Forex Decision Making Using Machine Learning
Friday I.K., Nayak D.S.K., Panigrahi R.R., Gochhait S., Swarnkar T.
Conference paper, Lecture Notes in Electrical Engineering, 2024, DOI Link
View abstract ⏷
The ever-evolving landscape of foreign exchange (FX) trading has seen a surge in the application of artificial intelligence (AI) techniques. In this fast-paced domain, fueled by ever-advancing technology, AI empowers traders to unlock new opportunities by harnessing the vast ocean of available trading data. This has led to the development of robust prediction models that enhance traders’ chances of success in the market. Traditionally, crafting accurate FX prediction models demanded expertise in a diverse range of fields, including quantitative analysis, financial literacy, and programming. However, the growing field of machine learning (ML) offers a more accessible and feasible alternative by leveraging trained algorithms to build remarkably accurate models. This study delves into a comprehensive review of various ML techniques deployed in the FX market, dissecting their diverse potential benefits. Additionally, it classifies the articles based on the types of ML model architectures, input indicators commonly employed, evaluation metrics, specific challenges associated with financial data and tasks, with potential solutions. This categorization aids researchers in staying abreast of cutting-edge advancements and facilitates replicating existing results for benchmarking purposes. Finally, the study concludes by illuminating promising avenues for future research within the research area.
ResNet-50: The Deep Networks for Automated Breast Cancer Classification using MR Images
Das T., Nayak D.S.K., Kar A., Jena L., Swarnkar T.
Conference paper, Proceedings of 2nd International Conference on Advancements in Smart, Secure and Intelligent Computing, ASSIC 2024, 2024, DOI Link
View abstract ⏷
One in every four cancer cases in women is breast cancer, which is the most prevalent malignancy in this group globally. In 2020, breast cancer will account for one in every eight new instances of cancer, according to projections of 2.3 million new cases. Breast cancer identification, both early and accurate, is critical to improve patient outcomes and survival rates. Magnetic resonance imaging (MRI) has developed as a valuable method for diagnosing breast cancer. Deep learning algorithms, notably convolutional neural networks (CNNs), have demonstrated exceptional effectiveness in a variety of medical image processing applications, including breast cancer classification. The ResNet-50 architecture has received a lot of interest in this context because of its excellent performance in image recognition tasks. ResNet-50 is a deep residual network that introduces skip connections and residual learning to solve the difficulty of training very deep neural networks. A total number of 1480 cancer MRI samples are collected from the Kaggle database, having two classes, which include healthy and malignant scans. It is observed that our proposed model with the data pre-processing techniques has achieved a classification accuracy of 92.01%.
Deep Learning Techniques for Identification of Pneumonia: A CNN Approach
Das R., Nayak D.S.K., Rout C.P., Jena L., Swarnkar T.
Conference paper, Proceedings of 2nd International Conference on Advancements in Smart, Secure and Intelligent Computing, ASSIC 2024, 2024, DOI Link
View abstract ⏷
Pneumonia is an inflammatory lung disease that mostly affects the tiny air sacs known as alveoli in human bodies. Pneumonia may be caused by bacteria, viruses, or fungi. Mild to severe symptoms may include a mucus-producing cough and a fever, appals, and difficulty to breath. The approach most usually used to diagnose pneumonia is chest X-ray (CXR) imaging. Images from a CXR screening can be used as an alternative or confirmation method because they are simple and affordable to acquire. Examining the chest X-rays however is a hard task that might be vulnerable to subjective variation. We used a deep learning technique in this investigation. Convolutional neural networks (CNNs) for deep learning were employed in this paper. In his work, the proposed model is applied to the original CXR image dataset and the augmented image dataset. We generated the augmented images from the existing CXR image dataset and deployed VGG-19, a component of the CNN model to create an autonomous Pneumonia detection system. The suggested techniques that have been employed include scaling the image, adding extra information, deep learning using Keras, batch normalization, and finally using weights from the pre-trained Keras model (VGG19). Finding pneumonia sickness is difficult and takes time. In this study, we address the issue and demonstrate how to detect pneumonia disease more quickly and accurately. Our proposed model shows an accuracy of 95% on the augmented CXR image dataset.
A Blockchain-Enhanced Framework for Efficient Paging and Energy Optimization in 5G Networks
Swain S., Meher S., Nayak D.S.K., Mohanty B.
Conference paper, 2nd International Conference on Signal Processing, Communication, Power and Embedded Systems, SCOPES 2024, 2024, DOI Link
View abstract ⏷
The rapid expansion of 5G networks (5GNs) has significantly increased resource demands, making energy efficiency a vital concern for sustainable operations. Key factors contributing to energy consumption include paging monitoring and Discontinuous Reception (DRX) optimization, which necessitate innovative solutions to meet the performance and energy-saving requirements of contemporary 5G systems. This study presents a blockchain-enhanced framework designed to address these challenges effectively. By leveraging blockchain technology's (BCT) decentralized and secure architecture, the framework utilizes smart contracts for real-time, automated monitoring and optimization. This approach minimizes the need for manual intervention, enhances transparency, and fosters trust in network operations. Transcending the limitations of traditional centralized methods enables dynamic and efficient resource management while preserving network responsiveness. Mathematical models illustrate how blockchain-driven DRX optimization and paging adjustments can reduce energy consumption without sacrificing latency or throughput. The robust capabilities of BCT enhance system reliability by ensuring data integrity and reducing operational overhead. This study underscores the transformative potential of blockchain in advancing energy-efficient, scalable, and secure 5G networks, providing a sustainable solution to the energy challenges within modern telecommunications.
HFai: Deep Networks for Detecting Heart Failure Progression
Conference paper, 3rd International Conference on Communication, Control, and Intelligent Systems, CCIS 2024, 2024, DOI Link
View abstract ⏷
Heart Failure affects millions worldwide, and early identification is crucial to prevent serious complications. Current diagnostic approaches depend mostly on human interpretation and are frequently inaccurate. Artificial Intelligence models are efficient for estimating heart failure progression since they provide healthcare professionals with more accurate and automated tools. This study aims to train and evaluate DL models (1D-CNN, DNN, ANN) on raw and feature-engineered cardiac arrest detection datasets. The performance of this model was first evaluated using heart disease data. On the raw dataset, ANN outperformed with the highest Accuracy of 80% and AUC of 82.74%. After feature engineering using XGBoost, ANN's performance increased with an Accuracy of 95.08% and an AUC of 93.99%, indicating that feature engineering improves prediction. These findings emphasize the potential based on AI models in developing more reliable and automated diagnosis techniques for cardiac diseases. Future research will integrate real-time patient monitoring data with improved predictive models to detect developing health issues and allow for more accurate early diagnosis. This technique can transform heart failure therapy by providing personalized and more effective medicines, improving clinical decision-making and patient outcomes.
Multichannel ECG-Based Heart Disease Detection with ResNet-50 and Decision-Tree Classifier
Conference paper, 2nd International Conference on Signal Processing, Communication, Power and Embedded Systems, SCOPES 2024, 2024, DOI Link
View abstract ⏷
The Electrocardiogram (ECG) is a non-invasive, cost-effective diagnostic technique for diagnosing cardiac diseases that monitors the heart's electrical activity. ECG signal interpreting is challenging due to noise, distortions, and patient variability resulting in delays, and errors in diagnosing heart disorders. The objective of this work is to improve the accuracy and reliability of ECG-based Arrhythmia identification through the use of advanced AI methods. The ResNet-50 (DL model) and Decision Tree (ML model) are used to enhance signal accuracy and identify heart abnormalities such as Normal (N), Left Bundle Branch Block (LBBB), Right Bundle Branch Block (RBBB), Premature, Premature Ventricular Contraction (PVC), Atrial Premature Beats (APB). The models are trained on the MIT-BIH Arrhythmia dataset, which includes Temporal, Wavelet Transform and Heart Rate Variability features to capture heartbeat patterns. The ResNet-50 model performs better than the Decision Tree in terms of Accuracy (99.71 %), Precision (99.08%), Recall (98.96%), Specificity (99.82%), F1-Score (99.01%), AUC (0.99), AUPRC (0.97). The proposed models showed potential for improving ECG-based cardiac diagnostics by enabling the efficient and accurate automated identification of heart irregularities.
Cybersecurity Challenges in IoT-Based Healthcare Systems A Survey
Book chapter, Intelligent Security Solutions for Cyber-Physical Systems, 2024, DOI Link
View abstract ⏷
Modern healthcare has undergone a digital transition driven by the Internet of Things (IoT). Due to the rise of chronic diseases and global aging in all nations, it is incredibly challenging for hospitals to facilitate and monitor every patient. Medical professionals and patients’ families are now able to continuously monitor patients’ conditions thanks to the use of wireless sensor networks for IoT-based medical systems. This is done by collecting data from sensors that are attached to patients. Security issues need to be taken into account from the outset of any digital transition. Due to the sensitivity of healthcare data, every breach endangers patient privacy. IoT networks are even more vulnerable to intrusion. Cyberattacks have the potential for fatal consequences. This study presents a survey and analysis of the literature on the use of IoT as a cybersecurity tool to safeguard IoT networks and proposes an assessment of anonymous authentication used in healthcare. The literature from 2017 to 2022 is thoroughly examined. A comprehensive examination of linked publications identifies a particular niche opportunity for researchers to pursue in this field.
Radiologic Decision-Making for Imaging in Pulmonary Embolism: Accuracy and Reliability of Large Language Models—Bing, Claude, ChatGPT, and Perplexity
Sarangi P.K., Datta S., Swarup M.S., Panda S., Kumar Nayak D.S., Malik A., Datta A., Mondal H.
Article, Indian Journal of Radiology and Imaging, 2024, DOI Link
View abstract ⏷
Background Artificial intelligence chatbots have demonstrated potential to enhance clinical decision-making and streamline health care workflows, potentially alleviating administrative burdens. However, the contribution of AI chatbots to radiologic decision-making for clinical scenarios remains insufficiently explored. This study evaluates the accuracy and reliability of four prominent Large Language Models (LLMs)—Microsoft Bing, Claude, ChatGPT 3.5, and Perplexity—in offering clinical decision support for initial imaging for suspected pulmonary embolism (PE). Methods Open-ended (OE) and select-all-that-apply (SATA) questions were crafted, covering four variants of case scenarios of PE in-line with the American College of Radiology Appropriateness Criteria. These questions were presented to the LLMs by three radiologists from diverse geographical regions and setups. The responses were evaluated based on established scoring criteria, with a maximum achievable score of 2 points for OE responses and 1 point for each correct answer in SATA questions. To enable comparative analysis, scores were normalized (score divided by the maximum achievable score). Result In OE questions, Perplexity achieved the highest accuracy (0.83), while Claude had the lowest (0.58), with Bing and ChatGPT each scoring 0.75. For SATA questions, Bing led with an accuracy of 0.96, Perplexity was the lowest at 0.56, and both Claude and ChatGPT scored 0.6. Overall, OE questions saw higher scores (0.73) compared to SATA (0.68). There is poor agreement among radiologists’ scores for OE (Intraclass Correlation Coefficient [ICC] = -0.067, p = 0.54), while there is strong agreement for SATA (ICC = 0.875, p < 0.001). Conclusion The study revealed variations in accuracy across LLMs for both OE and SATA questions. Perplexity showed superior performance in OE questions, while Bing excelled in SATA questions. OE queries yielded better overall results. The current inconsistencies in LLM accuracy highlight the importance of further refinement before these tools can be reliably integrated into clinical practice, with a need for additional LLM fine-tuning and judicious selection by radiologists to achieve consistent and reliable support for decision-making.
Genomic characterization and distribution of XDR Klebsiella pneumoniae ST15 carrying blaOXA-181 on ColKP3 plasmid from an ICH patient in eastern India: A global comparative analysis
Sahoo S., Routray S.P., Mohanty J.N., Pattnaik A., Nayak D.S.K., Shah S., Das J., Subudhi E., Swarnkar T.
Article, Gene Reports, 2024, DOI Link
DnnARs: An Artificial Intelligence Technique for Prediction of Antimicrobial Resistant Strains in E. coli Bacteria Causing Urine Tract Infection
Nayak D.S.K., Priyadarshini A., Mahanta P., Das T., Sahoo S.K., Swarnkar T.
Article, SN Computer Science, 2024, DOI Link
View abstract ⏷
The rise of antimicrobial resistance (AMR) in bacteria that cause infectious diseases poses a significant global health challenge, necessitating advanced prediction technology for effective management and response. This study aims to evaluate the predictive capabilities of several machine learning (ML) and deep learning (DL) models in identifying patterns of antibiotic resistance. We have proposed a comprehensive experimental framework that combines multiple machine learning algorithms and deep learning architectures to evaluate their efficacy and accuracy in predicting antibiotic resistance. We have developed DnnARs, an artificial intelligence pipeline designed to identify antibiotic-resistant strains of E. coli bacteria, which are mostly responsible for urinary tract infections (UTIs). DnnARs utilizes Synthetic Minority Over-sampling Technique (SMOTE) and applies feature engineering using principal component analysis (PCA) and t-distributed Stochastic neighbor embedding (t-SNE) on our dataset to evaluate the effectiveness of various AI models. Our model comparison demonstrates that DL models outperform ML models when the dataset is fine-tuned using SMOTE and feature engineering. Our DnnARs achieve a classification accuracy of 97%, a specificity of 91%, and a remarkable sensitivity of 100%. The DL models that were examined exhibit an average classification accuracy that is 3.66% higher than that of the ML models. The findings demonstrate the capability of deep learning methodology to enhance predictive analytics for antimicrobial resistance, offering a robust and efficient alternative to traditional machine learning techniques. This research aims to enhance diagnostic tools and medications, improve the management of infectious diseases, reduce antibiotic resistance, and integrate deep learning models into clinical workflows to enable immediate decision-making.
ARGai: AI Fusion Approaches for Classification of S. aureus Strains and Identification of Antibiotic Resistant Genes
Nayak D.S.K., Dash A., Routray S.P., Sahoo S., Swarnkar T.
Conference paper, Proceedings of the 2024 International Conference on Artificial Intelligence and Emerging Technology, Global AI Summit 2024, 2024, DOI Link
View abstract ⏷
Staphylococcus aureus (S. aureus) causes bone and joint infections, pneumonia, and bloodstream infections, the most common and fatal of these illnesses. Rapid identification of resistant strains and the antibiotic resistance genes (ARGs) linked to them can enhance the diagnostic process for various infectious illnesses. We introduce a new clinical support system that uses a hybrid multi-stage significant gene identification technique called MRMR-LASSO to select resistant and significant genes from next-generation sequencing (NGS) datasets. It is necessary to identify the S. aureus resistance strain from among thousands of relevant genes in NGS datasets, which are typically complex and comprise many genes with few samples. The curse of dimensionality makes it hard to pick informative genes from many genes. Optimization techniques are employed to resolve the gene selection issue. To select important genes for our ARGai model, we employed a filter in conjunction with an embedding approach. With its combination of machine learning (ML) and deep learning (DL) methods, ARGai can efficiently and accurately identify the important genes in S. aureus with a 96% success rate. Examining the biological relevance of chosen gene activity guarantees that ARGai's classification performance on susceptible and resistant strains is outstanding. Several cutting-edge AI models used for gene selection and strain classification are compared to our ARGai model. By shortening the screening period, this proposed ARGai may aid in subsequent phases of illness treatment, such as disease diagnosis and medication design.
Enhancing Fault Tolerance and Load Balancing in Cloud computing for improved e-healthcare Systems Performance
Priyadarshini A., Pradhan S.K., Laha S.R., Kumar Nayak D.S.
Conference paper, 2023 2nd International Conference on Ambient Intelligence in Health Care, ICAIHC 2023, 2023, DOI Link
View abstract ⏷
Cloud Computing is increasingly being recognized as a valuable virtual resource-sharing solution. Due to its ability to provide solutions across different industries, the number of Cloud computing users has grown substantially. Load balancing distributes the workload across multiple servers to optimize resource utilization and prevent overload, while fault tolerance ensures that the system can continue to function even in the event of failures. In an e-healthcare system, where downtime or errors can have severe consequences for patients, fault tolerance and load balancing become even more critical. Load-balancing techniques are of utmost importance in enhancing the effectiveness of cloud servers. We have comprehensively examined the primary elements that impact fault tolerance in load-balancing task migration within cloud computing. These factors have been assessed by employing various methods to evaluate their influence on fault tolerance in load balancing. By implementing these techniques, e-healthcare systems can improve their ability to handle traffic surges and ensure patient safety and satisfaction.
En-MinWhale: An Ensemble Approach Based on MRMR and Whale Optimization for Cancer Diagnosis
Panigrahi A., Pati A., Sahu B., Das M.N., Nayak D.S.K., Sahoo G., Kant S.
Article, IEEE Access, 2023, DOI Link
View abstract ⏷
According to the WHO, Cancer is a prominent cause of mortality worldwide, accounting for 10 million fatalities at the end of 2020. The most common types of cancers include Lung, Breast, CNS, Leukemia, Colon, and Cervical Cancer. Early detection of cancer can decrease the death toll. According to the study, if the cancer is identified at its early stage, the death rate can be reduced to 85%. In order to reduce the death toll, machine learning (ML) emerges as a significant solution. When it comes to cancer research with ML, biopsy and microarray data come into the front. The biopsy data is less useful as it excludes a patient's genetic information. However, due to the genetic information, the microarray data emerges as a solution to detecting cancer disease. Dealing with microarray data also has some consequences, and high dimensionality is one of them. This article reports an ML-based ensemble model to tackle microarray data issues and provide an effective model for cancer detection. The reported model uses Minimum Redundancy Maximum Relevance (MRMR) as the feature selection algorithm. The Whale Optimization Algorithm (WOA) is implemented for the featured dataset to select the optimistic number of features without affecting microarray data relevance. Then, four ML-based classification models, including Support Vector Machine, Decision Tree, Multilayer Perceptron, and Random Forest, are applied as the base learners to make initial predictions. Finally, the Voting ensemble technique is applied to the initial prediction to develop an ensemble model for cancer prediction. The proposed En-MinWhale model is evaluated over six different types of cancer microarray datasets, including Lung, Leukemia, Ovarian, CNS, Breast, and Colon Cancer. Finally, the performance of the proposed model is evaluated using 11 various evaluative parameters, including accuracy, precision, specificity, sensitivity, F- β score, etc. The En-MinWhale model shows 94.09%, 95.83%, 94.86%, 95.00%, 94.85%, and 96.77% accuracy for Lung, Leukemia, Ovarian, CNS, Breast, and Colon Cancer datasets, respectively, that outperforms other considered hybrid models and can help out the physicians in Cancer diagnosis.
Efficient Technique for Glioblastoma Multiforme Detection: Deep Learning or Transfer Learning
Mohapatra S., Kumari S., Kabi S., Sahu R., Kumar Nayak D.S., Swarnkar T.
Conference paper, IEEE Region 10 Humanitarian Technology Conference, R10-HTC, 2023, DOI Link
View abstract ⏷
The article discusses the challenges associated with the diagnosis of glioblastoma multiforme (GBM), which is a life-threatening form of cancer with low median survival rates. Medical image analysis is a challenge due to limited annotated data. However, transfer learning can help overcome this issue. In this study, deep learning models are trained to develop an automatic GBM diagnosis system that classifies the brain images as no tumor and glioma tumor. The performance of a customized convolutional neural network is compared to pretrained ResNet-50 and VGG-16 models based on metrics namely accuracy, specificity, sensitivity, precision, F1-score, and Matthews Correlation Coefficient (MCC). A critical analysis is conducted to determine the most efficient learning model for GBM detection based on the evaluated performance metrics. The study provides insights into developing a more efficient GBM diagnosis system, which can help in the development of novel therapeutic targets and treatment modalities for cancer.
An IOT-Based Soil Moisture Management System for Precision Agriculture: Real-Time Monitoring and Automated Irrigation Control
Laha S.R., Pattanayak B.K., Pattnaik S., Mishra D., Kumar Nayak D.S., Dash B.B.
Conference paper, Proceedings of the 4th International Conference on Smart Electronics and Communication, ICOSEC 2023, 2023, DOI Link
View abstract ⏷
Managing soil moisture is crucial for optimizing crop growth and yield in agriculture. With the advancements in Internet of Things (IoT) technology, innovative soil moisture monitoring and control solutions have emerged. This paper presents a pioneering IoT -based soil moisture management system tailored for precision agriculture within Siksha 'O' Anusandhan (SOA) University. The system utilizes sensor nodes, wireless communication, and cloud computing to enable real-time monitoring and automated irrigation control, addressing the limitations of conventional manual approaches. Providing continuous, remote, and precise measurements of soil moisture levels revolutionizes how soil moisture is managed within the university's agricultural practices. Low-cost sensor nodes are strategically placed throughout the fields to collect data, which is wirelessly transmitted to a central hub for analysis and decision-making. A cloud-based platform processes the data, generating actionable insights and triggering automated irrigation when necessary. Extensive testing and validation demonstrate the system's reliability and ability to improve water use efficiency, surpassing traditional methods. Furthermore, the IoT architecture of the system offers scalability and flexibility, making it adaptable to different crop types and field conditions present within SOA University. This innovative research contributes to the field of precision agriculture by providing an efficient and tailored solution for soil moisture management within the university's agricultural practices. This IoT -based system enhances crop productivity by enabling real-time monitoring and automated irrigation control, promoting sustainable water resource utilization. Its implementation within SOA University fosters the integration of cutting-edge technology into agricultural practices, facilitating optimized crop growth and contributing to the university's commitment to environmental sustainability and resource management.
ANTIBIOLOG: An Advanced Tool for Combat Antibiotic Resistance With Enhanced Multinomial Logistic Regression
Routray S.P., Nayak A.D., Nayak D.S.K., Sahoo S., Swarnkar T.
Conference paper, Proceedings of 2023 IEEE 9th International Women in Engineering (WIE) Conference on Electrical and Computer Engineering, WIECON-ECE 2023, 2023, DOI Link
View abstract ⏷
Pseudomonas aeruginosa, a multidrug-resistant organism with clinical significance, complicates the treatment of critically ill patients. Since traditional susceptibility testing takes at least 24 hours, timely identification and treatment are crucial in the face of rising Antimicrobial Resistance (AMR). In this study, we present the development of ANTIBIOLOG, a cutting-edge Machine Learning (ML) model. This model predicts P. aeruginosa antibiotic susceptibility based on patient demographics and antibiotic susceptibility testing results, using a dataset of 49 patient samples from IMS & SUM Hospital. Multinomial logistic regression was selected for dealing with multiple classes and providing probabilistic interpretations. The performance of the model is evaluated using various metrics such as accuracy, precision, recall, specificity, F1 score, and MCC. Through a user-friendly web interface, it simplifies resistance prediction and guides empirical antibiotic treatment. Unlike generic tools, it focuses on P. aeruginosa and provides accurate therapy selection. ANTIBIOLOG predicted antibiotic susceptibility with a promising 77.77% accuracy. This ML-based tool enables healthcare professionals to combat multidrug-resistant P. aeruginosa infections and address AMR issues.
A Comparative Analysis of Machine Learning Models for Colon Cancer Classification
Swain R.R., Nayak D.S.K., Swarnkar T.
Conference paper, 2023 International Conference in Advances in Power, Signal, and Information Technology, APSIT 2023, 2023, DOI Link
View abstract ⏷
One of the trickiest machine learning challenges is the automatic classification of cancer cells. Colon Cancer can strike anyone at any age, but it often strikes older persons. In recent years, several investigations have focused on developing machine learning algorithms that can automatically classify genes of various types of a particular disease, especially colon cancer. The classification of cancer has improved due to improvements in computational methods like machine learning (ML), which required little human interaction and inefficient computers. Since the classification of gene expression data has been extensively studied in conjunction with the advancement of gene technology, a wide variety of techniques, primarily neural network-related, have been applied to the analysis of medical data, which is primarily concerned with the large dimension and very small in quantity. In this work, five different ML techniques were implemented for the classification of colon cancer in different age groups. The implemented ML techniques are Random Forest (RF), Decision Tree (DT), Support vector Machine (SVM), Naive Bayes (NB), and K-Nearest Neighbour (KNN). It is observed that most of the implemented model has an average accuracy of more than 95%.
ReCuRandom: A hybrid machine learning model for significant gene identification
Nayak D.S.K., Pati A., Panigrahi A., Sahoo S., Swarnkar T.
Conference paper, AIP Conference Proceedings, 2023, DOI Link
View abstract ⏷
The method of evaluating the physical traits of living creatures by generating essential proteins is known as gene expression. Translation and transcription are the two processes in gene expression. Gene Expression can be captured from DNA or RNA using a variety of approaches. High-throughput gene expression (NGS) data is having multiple advantages over the traditional gene expression dataset. Still one of the major disadvantages of NGS data is it is complex, comprising a large number of genes with few samples; among these thousands of significant genes, disease-causing genes must be identified. This issue should be solved by lowering the data source's dimension to a manageable level. Machine learning (ML) has proven its potential in the research areas of disease identification, and genomics in recent years. Many Machine Learning-based Gene Selection techniques have been described in the literature, to improve dimensionality reduction precision. In this work, a unique clinical support system based on the Relief-Cuckoo Search (CS)-Random Forest (RF) hybrid multi-stage significant gene identification technique is proposed for significant gene selection from the next-generation sequencing (NGS) dataset. This work can be helpful for the accurate identification of sepsis and the relevant genes associated with it, which helps early detection of the disease. This hybrid feature selection technique could be beneficial in later stages of disease treatment, such as disease diagnosis and drug development. The accuracy of the proposed model is 95.23 % which can be helpful for further analysis of various infectious diseases and thus helps the diagnosis procedure more convenient.
A neural network framework for predicting Adenocarcinoma cancer using high-throughput gene expression data
Mohanty S., Nayak D.S.K., Swarnkar T.
Conference paper, AIP Conference Proceedings, 2023, DOI Link
View abstract ⏷
Machine learning is a part of artificial intelligence and its optimization techniques are allows system to learn from past example and identify pattern from large dataset. This criteria are well suited for researchers for doing their research on medical application, genomic measurements and it cause to machine learning techniques are used in cancer diagnosis and detection. Recent year machine learning has been applied to cancer prediction and prognosis. This latter approach growing trend towards an artificial neural network. ANN ensemble is a learning platform in which a number of neurons work together to solve a problem. Our approach is used to identify adenocarcinoma cancer diseases from gene expression data samples obtained from bodies that are to be diagnosed in this research. Last decades Next-generation sequencing (NGS) technology has been expanded with advantages in the reliability, data interpretation and costs, sequencing chemistry. Such large improvements can make NGS clinically practical, and NGS technology is presently being developed for use in the field of oncolo-gy. Various clinical applications are discussed, including DNA-sequencing-based mutation detection in inherited cancer syndromes, RNA-sequencing-based detection of spliceogenic variants, DNA-sequencing-based risk modi-fiers and application for pre-implantation genetic diagnosis, and cancer somatic mutation analysis. In this study, neural network classifiers were used to determine patient carcinoma cancer status. Applied neural network architecture is based on a two-level architecture, with each individual network having two outputs based on the normal cell sample or cancer cell sample according to our dataset.
Radiological Differential Diagnoses Based on Cardiovascular and Thoracic Imaging Patterns: Perspectives of Four Large Language Models
Sarangi P.K., Irodi A., Panda S., Nayak D.S.K., Mondal H.
Article, Indian Journal of Radiology and Imaging, 2023, DOI Link
View abstract ⏷
Background Differential diagnosis in radiology is a critical aspect of clinical decision-making. Radiologists in the early stages may find difficulties in listing the differential diagnosis from image patterns. In this context, the emergence of large language models (LLMs) has introduced new opportunities as these models have the capacity to access and contextualize extensive information from text-based input. Objective The objective of this study was to explore the utility of four LLMs-ChatGPT3.5, Google Bard, Microsoft Bing, and Perplexity-in providing most important differential diagnoses of cardiovascular and thoracic imaging patterns. Methods We selected 15 unique cardiovascular (n = 5) and thoracic (n = 10) imaging patterns. We asked each model to generate top 5 most important differential diagnoses for every pattern. Concurrently, a panel of two cardiothoracic radiologists independently identified top 5 differentials for each case and came to consensus when discrepancies occurred. We checked the concordance and acceptance of LLM-generated differentials with the consensus differential diagnosis. Categorical variables were compared by binomial, chi-squared, or Fisher's exact test. Results A total of 15 cases with five differentials generated a total of 75 items to analyze. The highest level of concordance was observed for diagnoses provided by Perplexity (66.67%), followed by ChatGPT (65.33%) and Bing (62.67%). The lowest score was for Bard with 45.33% of concordance with expert consensus. The acceptance rate was highest for Perplexity (90.67%), followed by Bing (89.33%) and ChatGPT (85.33%). The lowest acceptance rate was for Bard (69.33%). Conclusion Four LLMs-ChatGPT3.5, Google Bard, Microsoft Bing, and Perplexity-generated differential diagnoses had high level of acceptance but relatively lower concordance. There were significant differences in acceptance and concordance among the LLMs. Hence, it is important to carefully select the suitable model for usage in patient care or in medical education.
Prediction of Smoking Addiction Among Youths Using Elastic Net and KNN: A Machine Learning Approach
Conference paper, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2022, DOI Link
View abstract ⏷
In the current generation, it has been studied that smoking addiction among the youths is increased exponentially. Since there is a lot of awareness among the people about tobacco use but the youths are exposed to a broader spectrum of the different types of nicotine products like E-cigarettes, or hookah or water pipes, or conventional cigarettes, or dissolvable tobacco, and many more products. Our study aims are to identify some of the demographic factors like age, gender, sex, etc. to predict smoking addiction among the youths of our society. To predict e-cigarette addiction among youths, we considered the wings of Artificial Intelligence (AI) like Machine Learning (ML), and Deep Learning (DL) for better understanding and finding the relationship among the various features. During the data preprocessing we used the elastic net regression technique for feature selection and K-Nearest neighbor for making the accurate predictions. The Hybrid Prediction model was built by the Elastic net regression technique and KNN. It is observed that Elastic net finds a better selection of significant features. The outcome of the suggested pipeline provides high performance on the selection of significant features for the prediction model and provides better accuracy.
Quality Control Pipeline for Next Generation Sequencing Data Analysis
Conference paper, Smart Innovation, Systems and Technologies, 2022, DOI Link
View abstract ⏷
The current trends in biomedical research especially genome sequence analysis are enriched by the introduction of the high-throughput sequencing technique. High-throughput sequencing like next generation sequencing (NGS) plays a crucial role in the area of whole genome sequencing (WSG), epigenomics, transcriptomics, and many more. NGS data give a deeper insight into the results of various microbial and cancer genomics analysis. NGS data are focusing on quality and diversity over the traditional biological datasets. NGS data make the process of classification, clustering, and analysis more reliable with the help of advanced computational intelligence techniques like machine learning (ML) and deep learning (DL). The accuracy of the result mostly depends upon the dataset standard, similarly, the analysis of data is more effective when it is in normalized form. In this paper, we prepare a pipeline for preprocessing of NGS data of oral cancer and breast cancer into a normalized form and make them analysis-ready. The obtained result is in a standard format with normalized values. The normalized values of NGS dataset provide an ultra-edge dimension to the analysis process and help to obtain more accurate results.
An Enhanced Web-based Tools for Multiple Sequence Alignment: A Comparative Approach
Sahoo S., Routray S.P., Nayak D.S.K., Swarnkar T.
Conference paper, ICAN 2022 - 3rd International Conference on Computing, Analytics and Networks - Proceedings, 2022, DOI Link
View abstract ⏷
Multiple sequence alignment (MSA) is one of the oldest dilemmas in computational biology, involving the alignment of more than two DNA, RNA, or protein sequences. There are numerous methods for aligning multiple sequences. According to the survey, there are 103 free MSA Tools - Software, and Resources available. Many techniques are designed to maximize speed while ignoring alignment accuracy. Similarly, many techniques prioritize accuracy over speed. The MSA tool's goal is to reduce execution time and memory requirements while increasing the accuracy of MSA on large-scale data. Choosing which tools to use is one of the most difficult problems for bioinformaticians, especially for the majority of non-bioinformaticians. As a result, a comparison study of the various multiple sequence alignment tools is required so that biologists and bioinformaticians can correctly interpret their biological data using the appropriate tools. In order to determine which tool is best for multiple sequence alignment, we used a case study to evaluate various MSA tools.
IRGM: An Integrated RNN-GRU Model for Stock Market Price Prediction
Friday I.K., Godslove J.F., Nayak D.S.K., Prusty S.
Conference paper, Proceedings - 2022 International Conference on Machine Learning, Computer Systems and Security, MLCSS 2022, 2022, DOI Link
View abstract ⏷
Stock market price prediction has been very challenging because of the non-linearity of such a dataset. Several recent research studies have demonstrated the value of accurate prediction as a motivator for investors, researchers, and market analysts, regardless of market trend. Traditional approaches such as Linear Regression (LR) and Support Vector Machines (SVM) have produced results that are inefficient and inaccurate, leading to the development of Deep Learning-based solutions (DL). With the increased use of DL techniques in prediction, its performance is showing significant improvements over traditional time series models. With their ability to learn patterns and predict future prices, DL models such as Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN) are becoming increasingly popular. Even so, these models continue to evolve and are continually revised to make them more efficient and better. To forecast the price for the 11th day, we present an RNN-GRU (Gated Recurrent Unit) model based on four independent features and the prices from the previous ten days. Furthermore, this model's performance was compared to that of an LSTM and CNN model. The results of the last test revealed that our proposed model outperformed the CNN and LSTM models.
Deep learning approaches for high dimension cancer microarray data feature prediction: A review
Nayak D.S.K., Mohapatra S., Al-Dabass D., Swarnkar T.
Book chapter, Computational Intelligence in Cancer Diagnosis: Progress and Challenges, 2022, DOI Link
View abstract ⏷
Nowadays, high dimension microarray data are largely used in the field of biological research, critical disease prediction, and diagnosis. The microarray data contain a large number of highly correlated gene expressions in which feature extraction is required for accurate identification of informative genes and the disease type and thus makes diagnosis easier. The feature extraction of such large data is a complex task, and machine learning uses hand-crafted methods for feature extraction. But, by adopting the deep learning technique, it can be possible to infer the important features of genes automatically in microarray data by using various computational tools and methods. Gene expression profiling is very much required to specify the cellular condition concerning critical diseases and medicine development. For a critical disease like cancer, the cancer microarray data have a limited number of samples with large gene expression and features. The expression profile is very large, which makes it a challenging task to classify and predict the feature of high dimension microarray data associated with cancer. Deep learning is the new promising methodology for feature extraction in microarray data and is widely used in the field of genomics, clinical genetics, cancer identification, and diagnosis. In literature, several deep learning methods for averaging, normalized initialization, model selection, and learning algorithms are adopted for prediction of the feature of high dimension microarray data. In this work, the authors have given a detailed description of the gene expression in digital form as well as microarray form and explained the use of it in disease diagnosis. Moreover, a broad literature study has been carried out related to the feature prediction of high dimension microarray cancer data using various deep learning techniques. Last, the work is concluded by giving a critical analysis with several challenges and future research scopes.
Predicting Pediatric Appendicitis using Ensemble Learning Techniques
Pati A., Panigrahi A., Nayak D.S.K., Sahoo G., Singh D.
Conference paper, Procedia Computer Science, 2022, DOI Link
View abstract ⏷
Managing appendicitis in children is still lacking international consensus. This is due to the lack of defined international pediatric appendicitis management recommendations and the unavailability of data-driven studies. In clinical practice, heuristic scoring methods are frequently employed to determine patient urgency. The use of ensemble techniques may improve prediction responses, which is the major goal of this study. This study used an ensemble of classifiers to predict three target variables in 430 children and adolescents aged 0-18 years. The ensemble approaches used in this work are majority voting and weighted averaging. In the Jupyter Python environment, these strategies' results were confirmed in terms of accuracy, misclassification rate, sensitivity, specificity, precision, f1-score, false negative and positive rates, and Mathew's correlation coefficient. Combining machine learning and deep learning can help in diagnosing and managing pediatric appendicitis, allowing for more tailored medical decisions.
A Comparative Study using Next Generation Sequencing Data and Machine Learning Approach for Crohn’s Disease (CD) Identification
Nayak D.S.K., Routray S.P., Sahooo S., Sahoo S.K., Swarnkar T.
Conference paper, Proceedings - 2022 International Conference on Machine Learning, Computer Systems and Security, MLCSS 2022, 2022, DOI Link
View abstract ⏷
Multi-Drug resistance organisms (MDRO) are a major threat to the world. Crohn's disease (CD) is a chronic inflammatory disorder caused by bacteria, which mostly affects the gastrointestinal tract and is complex and heterogeneous. Generally, antibiotics are prescribed to treat CD but, overuse or misuse of antibiotics may lead to AMR. The trait's genetic architecture is mainly unknown. One of the medical genetics' main goals is to reliably predict CD based on these genetic and environmental factors, which may further help in the research of Antimicrobial resistance (AMR). The traditional approach for identifying the important genes that are causing the disease is complex with less accuracy. This complexity forces the researchers to search for an alternative, mainly focused on developing a prediction model with more accuracy and cost-effectiveness. The advancement in computational intelligence especially machine learning (ML) can handle the process of identifying the significant genes of this trait in an efficient and more biological significant manner. In this work, we use the recent growth of ML methodologies to identify the significant genes associated with CD. The effect of quality control (QC), imputing, and coding methods on non-linear model results revealed that QC and imputation of missing genotypes can artificially boost scores. The linear approach is found with an accuracy average of 0.52 whereas the accuracy average obtained with the non-linear model is more than 0.60. In addition to this, advanced high-throughput sequencing techniques like next-generation sequencing (NGS) help to improve the quality of data for analysis and thus lead to a better result. The ML approach especially the non-linear methodology like the extra tree classifier (ET) performs extremely well to find the CD with less computational time and high accuracy.
Interpretation of Optimized Hyper Parameters in Associative Rule Learning using Eclat and Apriori
Mohapatra D., Tripathy J., Mohanty K.K., Nayak D.S.K.
Conference paper, Proceedings - 5th International Conference on Computing Methodologies and Communication, ICCMC 2021, 2021, DOI Link
View abstract ⏷
The classification of frequent item set is a vital problem in data mining. In recent area, where it deals with more outcome within limited time, there is a need of certain algorithm supported by faster computation. The aim of this research work is to generate a frequent item set so as to achieve a comparatively reduced time and effort. Frequent item set is a set of items that is often bought with another item, for example bread and butter. Here, two association rules mining algorithm, that is, Apriori and Eclat are used to predict the probability of subordinate items being bought if a prime item is obtained. Apriori algorithm works by picking the frequent individual items in a database and extending them to larger and larger sets whereas Eclat algorithm uses depth first search for discovering frequent item sets. In this paper, a dataset of transactions in a retail company is taken and corresponding support, confidence and lift is estimated. Here the relationship between these variables is explored. It is observed in Eclat algorithm that even after varying the minimum support the maximum and minimum support remains constant and using the Apriori algorithm that increasing the minimum lift value results in a gradual decrease in maximum support and small increase in minimum confidence values.
Gene selection and enrichment for microarray data—A comparative network based approach
Nayak D.S.K., Mahapatra S., Swarnkar T.
Conference paper, Advances in Intelligent Systems and Computing, 2018, DOI Link
View abstract ⏷
Gene selection plays a vital role in understanding the disease progression and further it helps in understanding the therapeutic targets. Most of the genes available in micro array data are not informative for a particular disease of interest. Study of functional analysis and interaction structure of genes plays a vital role in selecting genes associated to complex diseases. This work uses two different network based approaches for gene selection and compares the biological and statistical enrichment of selected genes. Functional modules in the gene expression data are obtained using Gene Correlation Network (GCN) and marker genes in the modules are identified using R package Weighted Gene Co- expression Analysis (WGCNA). WGCNA is considered to be one of the best methods for analysis of global GCN using a suitable threshold that leads to a network with scale free topology. The differentially co-expressed genes are then compared with the existing gene selection approach which integrates the selected co-expressed gene modules with protein-protein interaction (PPI) network. Observation shows that using PPI network which is generated using multitude of high throughput experiments and available in public data bases selects more disease specific genes in comparison to constructed GCN. The study shows that integrative network analysis to find genes may provide greater insight in underlying biological response.
