Accurate and real-time gait phase detection has become a major challenge in current intelligent prosthetic systems. The main reasons for this are latency limitations, changes in sensor data, and the lack of effective temporal modeling in existing approaches. Traditional methods such as CNN, LSTM, and Transformer based architectures face problems such as high computational complexity, delayed inference, or limited compatibility on edge devices. To overcome these limitations, in this research we proposed SwiftGaitNet, a low-latency Edge-AI framework that is specifically designed for real-time prosthetic gait phase detection. The proposed system consists of Force Sensitive Resistor (FSR) signals, Inertial Measurement Unit (IMU) data collected by an IoT based wearable sensing setup and foot pressure information coordinated by multimodal sensor fusion. Lightweight temporal attention, gated temporal convolution, adaptive residual fusion and efficient fully connected decision layers are used in this architecture, so that faster sequential learning is possible with less computational load. Compared to existing methods, the proposed model is specially optimized for edge inference, thus providing fast and reliable real-time predictions without the need for heavy processing. According to the results of the experimental evaluation, SwiftGaitNet demonstrated better performance than the traditional baseline models CNN, LSTM, Transformer and CNN-LSTM. This model achieved 97.6% accuracy, 97.6% precision, 97.6% recall, and 97.6% F1-score, evaluated using a subject-independent, 5-fold Leave-One-Subject-Out (LOSO) cross-validation protocol on a dataset of 2250 temporal windows collected from 50 walking trials across four gait-phase classes (Heel-Strike, Mid-Stance, Toe-Off, and Swing). Inference latency, averaging 12.1 ± 1.8 ms, was measured directly on the deployed ESP32 DevKit hardware using TensorFlow Lite Micro rather than estimated offline. The proposed framework demonstrates computational efficiency while maintaining accurate real-time gait-phase detection. These results suggest that SwiftGaitNet is a promising candidate for low-latency, edge-deployed gait phase detection in wearable and prosthetic sensing systems; however, further work involving closed-loop prosthetic-control integration and validation with prosthetic users is needed before its suitability for clinical or real-time prosthetic-assist deployment can be established.

Banana cultivation plays an important role in the agricultural economy of Andhra Pradesh, particularly in the Ravulapalem region, which is well known for banana production. This study presents EAAF-Net (Ensemble AI Annotation Framework), an applied AI workflow for banana lifecycle classification and semi-automated image annotation using images acquired with a Raspberry Pi-based imaging system under controlled postharvest storage conditions. The workflow is designed to support the classification of three banana lifecycle stages—unripe, ripe, and spoiled—while reducing manual annotation effort and facilitating the development of reliable agricultural image datasets. Rather than proposing a new deep learning architecture, EAAF-Net combines three established pretrained convolutional neural network backbones, namely, EfficientNet-B0, Xception, and MobileNetV2, through ensemble probability averaging. A confidence-guided decision mechanism automatically accepts high-confidence predictions while identifying uncertain samples for manual verification, thereby improving annotation consistency and supporting efficient semi-automated dataset construction. The workflow is implemented as an edge-artificial intelligence (AI) system in which image acquisition and model inference are performed locally, without cloud computing or network-based Internet of Things (IoT) connectivity. The evaluation demonstrates that the proposed workflow provides competitive classification performance while offering a practical and scalable framework for confidence-guided annotation of banana lifecycle images. The proposed approach provides a useful foundation for intelligent vision-based applications in postharvest quality assessment, banana lifecycle monitoring, and precision agriculture, with future work directed toward broader field validation and IoT-enabled agricultural monitoring systems.

Progressive and precise detection of peanut leaf diseases is vital for enhancing crop output, reducing pesticide overuse, and promoting sustainable agriculture. Recent Deep Learning (DL) approaches such as Convolutional Neural Networks (CNNs) and Vision Transformers (ViT) have demonstrated promising results; however, their performance is often constrained by limited labeled datasets, class imbalance, and insufficient feature robustness. To tackle these issues, this study introduces an innovative hybrid architecture that combines a Deep Belief Network (DBN) with a ViT for the automated classification of peanut leaf diseases. The DBN is employed as an unsupervised feature learning module to extract robust and discriminative representations from raw leaf images, effectively capturing underlying disease patterns even with limited labeled data. These learned features are subsequently fused with the ViT architecture, enabling enhanced global context modeling through multi-head self-attention. Extensive experimentation on a curated peanut leaf disease dataset demonstrates that the proposed DBN-ViT model achieves a classification accuracy of 98%, outperforming conventional CNN-based and standalone transformer models.

The rapid adoption of AI in healthcare has made the clinical functionality more convenient at the expense of exposing multifaceted cybersecurity risks which are difficult to detect using the traditional perimeter- based approach to security. Although explainable AI is fully deployed with intent to achieve transparency and regulation compliance, the prospects of explainable AI as a native security control have not been pursued. The chapter is a presentation of Explainability- as- a- Defense (XaaD) a framework which transforms the results of AI explanations into continuous behavioral fingerprints to perform proactive cybersecurity monitoring. A XaaD architecture that takes advantage of a layered architecture with explainability generation, drift detection, and human- in- the- loop validation is proposed, which is shown in the example of a clinical decision support, medical chatbots, IoMT analytics, and federated learning. XaaD re- imagines explainability as a post- hoc interpretability condition into an active form of defensive health care AI performance of trust.

Maritime safety is a critical concern in the Internet of Things (IoT) era, where numerous connected devices are deployed to monitor and manage various aspects of maritime operations. This paper proposes an Improved Grey Wolf Optimized Fuzzy Logic Classifier (IGWO-FLC) based routing path selection algorithm and an improved RC4 based secured key management scheme for enhancing maritime safety in IoT environments. The IGWO-FLC algorithm is employed to determine the optimal path for data transmission in a maritime IoT network. It combines the advantages of the Grey Wolf Optimization (GWO) algorithm and fuzzy logic to consider distance, signal strength, reliability, and security. Moreover, in maritime IoT networks, an Improved Rivest Cipher 4 (RC4) based secured key management scheme is proposed. The results obtained through simulation demonstrate that the proposed approach provides improved outcomes of 295.3 kbps throughput, 99.25% packet delivery ratio, 2.4% energy consumption and 0.28% routing overhead.

As user-created content on social media platforms like YouTube reaches a large scale, the issue of toxic comments has become a critical concern to the digital communication process and the overall mental health of the user. Traditional methods of moderation such as manual or keyword-based do not meet the volume and linguistic complexity of an online conversation. The current models also lead to this issue by using simplistic binary classification systems that cannot reflect the overlap and complexity of internet toxicity. To resolve these shortcomings, this paper introduces a contextualized Context-Aware Toxic Comment Analyzer that is trained on a fine-tuned BERT-based bidirectional transformer architecture. The framework, unlike the previous ones, uses semantic tokenization to obtain contextually rich word representations that can be accurately classified into six categories of toxicity: Toxic, Severe Toxic, Obscene, Threat, Insult and Identity Hate. It is trained and tested using a dataset of around 160,000 YouTube comments, and it contains an extensive text preprocessing pipeline and word cloud-based analysis of exploration. The outcomes of the experiments reveal that the model is robust and has a high level of generalization with a classification accuracy of 96.77, AUC of 0.9853, training loss of 0.05 and testing loss of 0.065. The results make the system a scalable and viable automated content moderation solution, which can help to build a safer online environment.

Accurate and interpretable employee performance evaluation is essential for effective organizational decisionmaking. This work proposes a novel explainable classification framework, termed Self-Adaptive Genetic Boosted eXplainable Framework (SAGB-XSHAP), which integrates eXtreme Gradient Boosting (XGB) with a self-adaptive genetic optimization strategy and SHAP (SHapley Additive exPlanations) - based explanation stability analysis. The proposed approach jointly evolves feature interaction structures and model hyperparameters while explicitly penalizing unstable feature attributions during optimization. Unlike conventional methods that treat interpretability as a post-processing step, the proposed framework embeds explanation consistency directly into the learning process. Experimental evaluation on a structured employee performance dataset demonstrates that the SAGBXSHAP framework accomplishes a classification accuracy of 98%, outperforming traditional Machine Learning (ML) and standard boosting-based approaches. Furthermore, SHAP-based analysis reveals consistent and meaningful influence patterns for key employee attributes such as salary, overtime, and job involvement, enabling transparent and reliable performance assessment. The results confirm that the SAGB-XSHAP framework precisely balances predictive accuracy with explanation stability, making it acceptable for practical human resource analytics implementations.

Social media has become an integral part of daily life, with platforms like Twitter serving as popular outlets for users to share information and express grievances. While social media offers numerous benefits, it can also be misused for cyberbullying, such as insults and harassment. This highlights the importance of detecting and mitigating cyberbullying to ensure a safe online environment. Detecting sarcasm in social media posts is challenging and requires advanced automated systems. To address this, a deep learning-based approach was developed using a Convolutional Neural Network (CNN) for analysis and an Attention Mechanism-based Bidirectional Long Short-Term Memory with Gated Recurrent Unit (AM-BLSTM-GRU) for prediction. Sarcasm detection datasets were gathered from sources like Kaggle and News Headlines. Standard NLP-based auxiliary features were extracted, and an embedded CNN model refined these features into feature vectors. The AM-BLSTM-GRU model then performed sarcasm detection and sentiment classification. The Enhanced Sinogramic Red Deer (ESRD) optimizer was utilized to optimize the classifier parameters effectively. The proposed model outperformed existing deep learning approaches on popular benchmark datasets and evaluation metrics, demonstrating its effectiveness in detecting cyberbullying. This approach was validated using well-known datasets and metrics, confirming its efficacy in identifying and reducing harmful online behaviors.

Face Recognition is one of the most advanced and drastically growing research areas because it helps identify people globally in various ethical and unethical applications. Face recognition needs face detection that can be compared with a list of available faces to predict the correct person. Face detection has become popular, easy, and fast since it follows the Viola-Jones FD method. Face comparison is obtained by comparing the internal and external information from the face images, like different features, face structure, key points, and patch-by-patch comparison. Earlier face recognition methods used separate algorithms for feature extraction from the face images, like color, shape, texture, histogram, and local and global binary pattern, to compare pairs of images where they provide more complexity regarding computation, cost, and time. After the evolution of artificial intelligence models, recent research has focused on using machine and deep learning algorithms for face detection and recognition. However, the accuracy of face recognition models needs to be improved under various conditions. Thus, this paper used a two-stage face comparison model to enhance face recognition efficiency. A consequence of three CNN models called CNN-1, CNN-2, and CNN-3 are used to detect the faces, detect the facial features, and recognize the faces, respectively. The CNN models are implemented in Python, and the results are verified by experimenting with multiple benchmark face datasets. The output accuracy obtained from the face detection and recognition is compared with the facial feature detection and recognition to choose the best to identify the criminals. From the comparison, both FDR and FFDR obtained 99.68% accuracy equally

The integration of Internet of Things (IoT) devices with blockchain technology offer enhanced data security, traceability, and decentralization. However, IoT devices face challenges such as limited computational power, energy constraints, and fluctuating network conditions, which make robust cryptographic implementation difficult. Digital signatures, which are essential for data integrity and authenticity in distributed blockchain systems, must be adapted to these constraints. While the Edwards-curve Digital Signature Algorithm (EdDSA) is secure and efficient, its static threshold implementations do not suit IoT networks, where rigid thresholds can cause excessive computational overhead or reduced security. Our work introduces a Dynamic Threshold EdDSA (DT-EdDSA) that adjusts signature thresholds in real time based on transaction priority, device capacity, and network conditions. By dynamically adapting the cryptographic intensity, DT-EdDSA reduces the computational and energy overhead on resource-limited devices while maintaining strong security during critical operations. Simulations show that DT-EdDSA significantly improves energy efficiency, signature generation time, and security resilience compared with static models, addressing IoT security limitations. This approach provides a flexible, context-aware cryptographic solution for IoT-blockchain networks, underscoring the relevance of adaptive techniques in meeting the security challenges of decentralized IoT environments.

Urban intersections face persistent challenges due to traffic congestion, which affects travel efficiency and increases delays. This study presents a Deep Reinforcement Learning (DRL)-based approach to optimize traffic signal timing within a SUMO (Simulation of Urban MObility) simulation environment. The suggested method focuses on minimizing vehicle delays, waiting times, and congestion through real-time adaptive control of signal phases. The system is modeled around a four-way intersection with distinct signals for straight and left-turn lanes. A Deep Q-Network (DQN) is trained using a reward mechanism that incentivizes smoother traffic flow and reduced wait times. The model leverages state representations derived from vehicle occupancy and distance from traffic lights, and incorporates experience replay for consistent learning. One key aspect of the system suggested is the prioritization of emergency vehicles like ambulances by dynamically modifying signal phases to provide immediate passage. Experimental results show drastic improvements over conventional methods with reduced queue length, higher throughput, and quicker emergency response. This research offers a scalable and resource-efficient solution for modern urban traffic management, with potential applicability to realworld scenarios involving dynamic and mixed traffic conditions.

In this fast-changing educational landscape of today, data science and predictive analytics are tools critical to creating student success and transforming educational systems. This chapter will further explore how predictive analytics can be utilized to anticipate and improve student outcomes. It also includes methodologies in collecting and analyzing student data, algorithms predicting their academic performance, and insights for early interventions and adapted support by educators and administrators. The predictive model, based on historical and real-time data, can predict the at-risk or chance of succeeding in student and develop learning paths for each one. The chapter also tackles data privacy issues, ethical implications, and the AI technology integration processes in schools. This chapter explains how predictive analytics the power can have to offer a better personalized, fair, and effective learning environment that would ensure improved student success and retention.

Skin infections are a major concern for human health, as they can cause significant skin damage, leading to loss of confidence and emotional distress in patients. Advancements in deep learning offer promising solutions for diagnosing and treating such conditions effectively. AI-driven approaches enable automated skin disease detection without requiring expert intervention, making diagnosis more accessible. Enhancing the user interface of these systems can further improve user experience. Early identification of skin disorders is crucial in preventing misdiagnosis as minor allergies, which can otherwise lead to severe complications. This research explores the application of deep learning for improved skin infection detection and treatment. Leveraging the power of AI, the study introduces a novel classifier combining the VGG-19 convolutional neural network with Grad-CAM (Gradient-weighted Class Activation Mapping). This approach aims to enhance diagnostic accuracy and reduce the risk of misdiagnosis, ultimately minimizing patient complications. The model was trained and evaluated using a dataset sourced from Kaggle, a popular platform for machine learning datasets. Performance was compared against baseline machine learning models, including decision trees and Support Vector Machines (SVMs). Results indicate that the proposed dual-input model, incorporating VGG-19 and Grad-CAM, achieved a remarkable accuracy of approximately 96%. This significantly outperforms the baseline models, demonstrating the potential of deep learning techniques for accurate and efficient skin condition diagnosis. The improved performance suggests that this approach could be a valuable tool for dermatologists and other medical professionals in the future.

The Internet of Medical Things (IoMT) is a revolutionary advancement in the healthcare field, allowing real-time monitoring and data gathering from medical devices attached to the human body. However, the sensitive nature of physiological data traveling over wireless networks makes IoMT networks extra sensitive to advanced cyber-attacks. Conventional IDS approaches often struggle with overfitting and fail to generalize well in dynamic attack scenarios, particularly in resource-constrained environments. To address this challenge, this study proposes a novel hybrid feature selection method, AQU-IMF-RFE, which integrates Aquila Optimization (AO) and Mutual Information (MI) with Recursive Feature Elimination (RFE). The methodology begins by calculating the mutual information scores for each feature to capture relevance, followed by optimization using the AO algorithm to handle redundancy and identify an optimal subset of features. Finally, RFE was applied to refine the selection through iterative feature ranking using machine learning feedback. This three-stage approach enhances the detection accuracy, reduces the dimensionality, and ensures computational feasibility. The method was tested on three benchmark IoMT datasets—UNSW-NB15, BOT-IOT, and CICIDS2017—with excellent classification accuracies of 94.35%, 94.48%, and 92.26%, respectively–demonstrating its effectiveness. These results not only outperform current state-of-the-art methods but also indicate the efficacy of the proposed method in identifying a wide variety of attacks. Thus, the proposed AQU-IMF-RFE framework represents a significant advancement in securing IoMT environments through intelligent and adaptive intrusion detection.

Diabetic Retinopathy (DR) is a prevalent eye condition among diabetic patients that can lead to irreversible vision loss if not detected and treated early. Early detection and continuous monitoring are crucial for effective management of DR. Traditional diagnostic methods, including visual acuity tests and non-invasive imaging, are often time-consuming and less efficient. Previous approaches, such as Residual Contrast Limited Adaptive Histogram Equalization (RCLAHE), have shown limitations in detecting DR. This research presents an Optimized Back Propagation based Deep Residual Learning Network (Op-BPDRLN) algorithm to enhance the detection of DR. By comparing the performance of different ensemble models on various retinal image datasets, this study aims to develop a more efficient and accurate algorithm for DR diagnosis. The comparison is based on key metrics such as accuracy, error, precision, and recall to determine the most effective classification model.

The most common catastrophes are rains, which cause fatalities and the destruction of livelihoods, both of which have an impact on the economics of the countries. Several research and innovative methods have been developed to effectively create Rainfall Forecasting Systems (RFS). The authors observed and discussed the current trend in rainfall prediction towards data-driven techniques. For predicting tasks, the ensemble Machine Learning (ML) and Deep Learning (DL)-based models that were trained using historical data on climatic parameters are becoming more and more valuable. The main goal of this work is to study ensemble ML and DL-based algorithms to illustrate recent advances in rainfall forecasting. The authors examined several well-known ensemble machine learning and deep learning methods for rainfall forecasting, which a variety of experts may utilize to create their solutions.

Colleges and universities place a premium on student job prospects since placements are often considered a measure of success. Globalization, digitalization, and new advancements in Artificial Intelligence (AI) are fascinatingly altering the job market landscape at a quicker rate than ever before. Any school would do well to familiarize itself with the current job market's demands and the most important elements influencing employability. Computer Science and Engineering (CSE) programs throughout the world are always looking for new ways to improve their curricula and better prepare their graduates for the workforce. One common method for looking into the future from the past is time-series analysis. Machine learning (ML) and fact-driven approaches are used in academic institutions. Predicting future career prospects using ML techniques is becoming an increasingly popular area of study. This research examines the potential for employment of CSE graduates taking part in campus hiring employing a time-based methodology. Based on three worldwide features- recession (g1), over-hiring (g2), and automation (g3)-this research used Facebook Prophet (FBP) and eXtreme Gradient Boost Algorithm (XGBA) to evaluate the career prospects of CSE graduates during campus hiring. The performance of FBP and XGBA was explored and assessed in terms of accuracy by considering global aspects into account. A 94.6% success rate was produced by the FBP model and XGBA after taking global variables into account. The research also found that major worldwide events, such as automation, recession, and overhiring during COVID-19, had consequences on CSE students' hiring process.

The goal of this project is to create a sign language identification system that uses computer vision and machine learning to help people who are deaf or hard of hearing communicate more effectively. This technology can considerably increase their capacity to speak and be understood by accurately recognising signs (perhaps using Support Vector Machines). This technology has a wide-ranging societal impact. It could be used in public settings, instructional materials, assistive equipment, and even for personal purposes. This sign language detecting technology has the potential to break down barriers and provide a more inclusive atmosphere in which everyone, regardless of communication ability, may engage in social interactions. This technology development offers a huge move towards a more compassionate and inclusive future for those with diverse communication requirements.

Multiple sclerosis (MS) is a complex autoimmune disease-causing damage to nerve coverings and disrupting communication between the brain and the body. Timely and accurate detection of individual responses to treatment is crucial for effective personalized MS therapy. Traditional models like logistic regression and decision trees offer some predictive accuracy, yet the variability in symptoms and treatment responses poses challenges. This research study suggests an innovative MS diagnosis approach, integrating machine learning (ML) techniques, specifically ensemble learning methods (random forest, bagging, stacking), and convolutional neural networks (CNNs). The complexity of MS and diverse patient responses necessitate sophisticated techniques. Ensemble learning enhances predictive capacity by capturing intricate relationships within heterogeneous data, including magnetic resonance imaging (MRI) and clinical records. CNNs, specialized in image analysis, contribute to spatial feature extraction from MRI scans, improving sensitivity to subtle MS-related changes.

Domain Name System (DNS) security refers to the steps taken to secure the confidentiality, integrity, and accessibility of DNS-related information in a blockchain-based system. DNS is a critical internet component that converts human-readable domain names into IP addresses, permitting users to access web pages and other online resources. Because every node in the network has cryptographic keys, it may access resources and take part in the consensus process, establishing an encrypted environment for DNS operations. We propose an innovative combination of web-based DNS management with a blockchain-based firewall in the realm of network management. The primary objective of project is to utilize blockchain technology to add an additional barrier of defense while strengthening the security, integrity, and efficiency of DNS operations. In addition, the proposed system has a blockchain-based firewall that adds a distributed mechanism for evaluating inbound DNS requests and intercepting malicious traffic.

Intrusion Detection Systems (IDS) play a crucial role in safeguarding modern digital infrastructures by identifying potential threats and anomalies in real time. As cyberattacks become more sophisticated, leveraging Machine Learning (ML) techniques in IDS has emerged as a promising approach to enhance detection accuracy, adaptability, and resilience. This paper provides an in-depth exploration of various ML methods applied to IDS, categorizing techniques such as supervised, unsupervised, and reinforcement learning. Additionally, it delves into the most used datasets for training and evaluating IDS models, highlighting their characteristics, advantages, and limitations. Furthermore, the paper addresses the persistent challenges in deploying ML-driven IDS, including issues related to data imbalance, real-time performance, adversarial attacks, and model generalization. Through a comprehensive analysis of current research and future directions, this survey aims to offer insights into the evolving landscape of ML-based IDS, paving the way for more robust and scalable solutions in the face of ever-evolving cyber threats.

India, renowned for its tomato production, faces challenges in terms of quality and quantity of tomato crops due to escalating incidences of diseases. This research introduces 'Tomato Guard,' a predictive model specifically designed to forecast diseases in tomato plants, aiming to mitigate these challenges. Utilizing advanced machine learning techniques and integrating diverse datasets comprising environmental variables, plant health metrics, and historical disease trends, the model adopts a proactive stance towards disease management. Rigorous field trials validate the efficacy of 'Tomato Guard,' showcasing its superior accuracy in early disease detection compared to traditional scouting methods. Its adaptability to various climates and geographical regions enhances its utility for growers, offering a versatile solution to combat disease outbreaks. Furthermore, the model not only minimizes economic losses but also advocates for sustainable farming practices, thereby contributing to global food security and agricultural resilience. The model's user-friendly interface facilitates interaction, enabling growers to upload diseased images and receive accurate predictions along with suitable remedies. Overall, 'Tomato Guard' stands as a significant advancement in disease prediction and management in tomato cultivation, empowering growers with actionable insights to safeguard their crops and sustain agricultural productivity.

One of the oldest techniques used in telecommunication for encoding regular characters is Morse Code. Morse Code is categorized into two separate electronic pulses which are dot (aka short pulse) and dash (aka long pulse). Detection of text from images of morse code is a complex process and there is no active research on this area As these are morse code images, different images have different styles of strokes. Our work aims to develop an Automated Morse code recognition system which is trained by a CNN (convolutional neural network) model with a self-built dataset and involves in collecting and preprocessing images of Morse code characters and creating a labeled dataset for training and testing the CNN model. The dataset creation process includes capturing images of different Morse code characters, augmenting the data to increase the dataset size, and annotating the images to label them correctly. The CNN model is then trained using the created dataset and evaluated for its accuracy in recognizing Morse code characters in images. The results demonstrate comparative analysis of different CNN based frameworks and achieved high accuracy in recognizing Morse code characters in images, making it a promising solution for automated Morse code recognition systems.

Amharic is made the official working language of Ethiopia from the late 12th century onwards. Amharic is a language with a total of 57.5 million speakers, out of which 32.4 million people are being the first-language speakers or native speakers and remaining are alternative-language speakers. Amharic is a distinctive syllabic writing system that employs 33 consonant symbols and each one's seven vowel variants. Due to the ambiguity among constant symbols, it is difficult to convert handwritten simple words in Amharic to English. This research study focuses on detecting the text in Amharic words in handwritten images and converting them to English text using various Convolution Neural Network Based frameworks such as AlexNet, Lenet-5, and ResNet-152 available today. For that, a new dataset has been created with more than 60, 000 training and 10000 testing images of Amharic constants written by various age groups and named it LBRAMHARIC. After that, various CNN-Based frameworks are trained with LBRAMHARIC and tested on the random inputs and customized input images. In the end, a comparison has been made on the obtained results by considering the different evaluation criteria such as accuracy and prediction rate.

In India, the number of crimes committed against women and girls is rising quickly every day. Even in our developed environment, women are unable to live autonomously. Domestic violence, assaults, threats, and teasing of women and girls are recorded often each day. Women are worshipped as goddesses in India. Women play many roles like mother, sister, and spouse and guide men in many aspects. But they are underrated and face unexpected assaults, domestic violence, and threats from society. One female in India is physically assaulted every 1 hour, as per the latest official data. But owing to a lack of sufficient proof, the offender escapes punishment. Such incidents are 99% unidentified, undetected, and undocumented. Surveillance cameras are somewhat helpful, but when the incident takes place in a secluded place, it is impossible to track down the suspect. One of the key causes is that school children are not taught the distinction between pleasant and unpleasant touch. Girls who are aware of the unpleasant touch but don't want to tell their parents tend to do so reluctantly or wait until after the crime has been done, which is useless. In such circumstances, the chance of the offender being punished is relatively minimal. Many researchers have suggested, designed, and implemented Women's Safety Devices (WSD) which protect them from unexpected abuses, assaults, teasing, and threats. There are many WSDs implemented by many researchers in this modern world using the Internet of Things (IoT), Embedded Systems, and Mobile apps. In this work, we studied the existing methods related to Embedded and IOT-based WSD that worked effectively and efficiently and presented challenges that most of the researchers faced while developing WSD. We hope these challenges and research gaps may be useful for building new, efficient, flawless, and accurate WSD by the existing and young researchers who are working in the domain of WSD.

Several approaches have been proposed in the published and existing scientific literature to assist farmers in decreasing the possibility of human error, but these approaches are expensive to execute since they rely on sensors and image-capturing equipment. Machine Learning's (ML) subfield of Deep Learning (DL) has changed the course of study toward how to classify epidemics. DL systems have very great potential for accurately identifying grape leaf diseases. In this work, different existing approaches that accurately identify grape leaf diseases were studied and analyzed. The biggest challenge of the current research is to examine and evaluate ML and DL methodologies that can effectively analyze grape plant diseases using leaf images. Leaves are considered reliable indicators of plant health, and this research aims to address the urgent national concern of enhancing agricultural productivity. The emergence of computerized identification and categorization of disease technologies for grape trees is still in its infancy, according to analysts. Farmers need new, better systems for identifying and classifying grape plant leaf ailments to avoid financial catastrophe and increase fruit harvest. This study presents a thorough systematic analysis of DL and ML techniques for identifying grape leaf disease. This study is expected to be useful for the field of agricultural disease research in the early detection and management of illnesses affecting grape leaves.

Farming is India's primary economic sector. More than 58% of rural income is generated by the agricultural sector. One of the most extensively produced food crops in India is tomatoes. So, as tomato plants grow less resistant to disease, early diagnosis is more important than ever. The plant's output will drop if proper maintenance is neglected. Tomato leaves are susceptible to a wide range of leaf-borne illnesses. When insects destroy crops and plant life, it slows a country's progress. To find and differentiate signs of plant sickness, experts often inspect the plants without the use of a microscope. The problem is that this approach is often inefficient, expensive, and time-consuming. A comprehensive bibliographic review of Deep Learning (DL) and Machine Learning (ML) methods for Tomato leaf disease identification is presented in this research. It is believed that this work will serve as a helpful tool for the agricultural disease research community using technology for early identification and control of tomato leaf diseases.

Citrus plants cannot develop quickly due to illnesses of the fruits and foliage. Therefore, creating a computer-based citrus fruit/leaf disease detection system is crucial. Artificial Intelligence (AI)-based computerizedcitrus fruit/leaf recognition is now possible because of Machine Learning (ML) and Deep Learning (DL) algorithms. Timely issue detection is essential for growers of citrus plants to avoid impeding their development or even preventing citrus plant mortality. It will be easier for agricultural workers to quickly remove citrus plant illnesses if the software is developed that autonomously recognizes them. In this study, a CNN-LSTM is used to create a detection model of citrus plant illnessesthat divides images of diseased citrus leaves into two categories: unhealthy and healthy. Further, hyperparameters like optimizers and Learning rates are adjusted to get maximum accuracy. The results are analyzed, interpreted, and tabulated in this work. The maximum accuracy of the CNN-LSTM model is obtained at a learning rate of 0.001 and by using the Adam optimizer. The maximum accuracy achieved by CNN-LSTM is 96.8% which is better than state-of-the-art approaches. Finally, it is concluded that the CNN-LSTM model is better at identifying citrus leaf diseases efficiently. The CNN-LSTM model is a successful method for differentiating between healthy and diseased citrus leaf images in terms of accuracy findings.

India lies in the tropical wet and dry area and gets a huge amount of rainwater in the monsoon season accounting for most of the yearly rainfall. Predicting the rainfall in these types of tropical areas is a difficult task and moreover, the influence of environmental issues has resulted in a substantial rise in the average rainfall distribution as well as the severity and frequency. In this work, for predicting the rainfall various Time series forecasting models like Seasonal Autoregressive Integrated Moving Average (SARIMA), Holt's Winter Seasonal Method (HWSM), Generalized Auto-Regressive Conditional Heteroskedasticity (GARCH), and Holt's Linear Trend Method (HLTM) have been used. Finally, a comparative analysis was made to inspect the best time-series forecasting model. The evaluation parameters used to perform a comparative analysis of different Time Series forecasting models are RMSE (Root Mean Square Error), Mean Absolute Error (MAE), and Mean Square Error (MSE). It was observed that the HWSM model had recorded the least error rate of 5.767 and 5.343 as RMSE, 2.143 and 2.432 as MAE, 31.65 and 26.75 as MSE for dataset-1 and dataset-2 respectively when compared to the remaining models.

The Arabic Language is the broadest and most used Language in the Arab World, with more than 423 million people. There are various representations in the Arabic Number system like Hindu-Arabic, Mashriqi Arabic, Eastern Arabic, and Western Arabic, which are look-a-like. Handwritten Character Recognition or Optical Character Recognition has gained popularity in recent days. This paper focuses on the importance of Optical Character Recognition (OCR) and different CNN-Based frameworks such as Lenet-5, ResNet-101, and ResNet-152 available today for OCR. This research study has created a dataset with more than 50,000 training and 7000 testing images of Arabic numbers ranging from 1 to 100 numbers written by various age groups and named it LBREADB. After that, various CNN-Based frameworks were trained with LBREADB and tested on the random inputs. A comparison has been made among the results based on the measurement criteria such as accuracy and prediction rate, which were shown at the ending.

Most all datasets that hold real-time data have an imbalanced organization of class instances. The total quantity of instances in certain classes is substantially greater than other classes and this skewed nature in the arrangement of classes is called Class Imbalance Problem (CIP). This imbalanced data affects the prediction performance since this forecast the weak class data samples wrongly. CIP is experienced by data mining professionals in a broad range of sectors. The categorization of imbalanced data is a huge challenge that arises in the discipline of Machine Learning (ML) and Deep Learning (DL). It is the critical issue that emerged for research and the deployment of sampling strategies to enhance the performance of the classifier has attracted extensive interest in the literature review. In this study, the importance of organizing imbalanced data is explained and the techniques suggested by the different scholars to counterbalance the skewed nature of classes and the assessment criteria for measuring the accuracy and prediction rate of the different classifiers have indeed been examined.

Rainfall in India is becoming more unpredictable, leaving it harder to forecast. When it comes to predicting Indian summer monsoon rainfall, the Indian Meteorological Department (IMD) now uses Ensemble methods and Statistical methods. As a result of this, strategists are unable to foresee the socioeconomic consequences of floods (too many rains) or droughts (fewer rains). Precisely how much rain falls depends on several variables such as a measure of the warmth or coldness in the air, moisture, breeze, movement, and direction of the wind. This article will use Ensemble time-series forecasting model ARIMA (Autoregressive Integrated Moving Average) + GARCH (Generalized Auto-Regressive Conditional Heteroskedasticity) to forecast the intensity of rainfall by considering various meteorological factors like sea-level pressure, moisture, dew point, min-max temperature, snowfall, geopotential height, speed and direction of the wind, humidity, and atmospheric pressure. The suggested Ensemble ARIMA+GARCH model has given good results when compared with individual models and state-of-the-art ensemble approaches in terms of RMSE, MAE, and MSE.

Telugu is a South Indian language that more than 10 Crore people spoke in this world. The alphabets of the Telugu language are more complex, alpha-syllabary, and agglutinative in nature. Different alphabets are like each other and will overlap with other alphabets. Due to its core, the handwritten alphabets in the images cannot be detected easily by the existing optical character recognition system. Our project will create a dataset of various representations of Telugu alphabets to avoid these problems. The proposed system can be trained to detect the character present in the images. This paper proposes a new technique to identify the handwritten alphabets of Telugu language in the pictures by using a Deep Convolutional Neural Network. A remarkable improvement has been achieved by recognizing the individual alphabets with an overall accuracy of 80% - 95% with the proposed model.

Blockchain will be the most used technology in the next decade with the key features of distributed ledger and high security. To provide security in the transactions process of hashing, blockchain technology was using the Elliptic Curve Digital Signature Algorithm (ECDSA). Using the ECDSA Algorithm, if any error has happened, this may lead to identify the private key value that which can acquire matching signatures for various documents. To avoid that problem, we suggest using the Edwards-curve Digital Signature Algorithm (EdDSA) Algorithm in generating the hash functions among the transactions, which provides the high-level speed, getting the best performance, and independence in generating the random number. Our suggestion provides improved security while compared to the ECDSA Algorithm.

Unmanned Aerial Vehicles (UAVs) are the most common applications used by many organizations in the present situation. Like every application, there are different issues and challenges with these UAVs also. One of the main cyber threats to UAVs is hijacking, which can be easy with small IoT devices. Also, they lack the privacy of the data which was captured by the UAVs, such as Images, Audio, and Video. This chapter discusses different challenges and threats that cause the tricky to Unmanned Aerial Vehicles, such as hijacking, privacy, cyber-security, and physical safety. The different solutions for avoiding those issues with current technologies are also discussed.

These days, usage of the web turned into a typical custom in each human's life. As innovation propels, once in a while it will prompt misuse. One of the abuses of innovation in the current circumstance is Deepfakes, which means producing trick recordings from the source video. In this Deepfakes, it enables the designer to change the substance of one individual in a video to someone else, changing the source voice with made voice, facial changes of the source in a sensible way dependent on the Artificial Intelligence ideas. Our fundamental point is that the client shouldn't accept these phony recordings which were populated on the web. For that reason, in this paper, we portray another strategy to find the Deepfakes online with the assistance of CART (Classification and Regression Trees). The fundamental preferred position of our technique is recognizing whether the video is phony or unique on the web, which encourages the client to realize whether to accept the substance of the video or not.

Women safety has turned into a major issue in today’s society. According to recent surveys, India stood top in insecurity for women among society, which explains that how women feel insecure in our country. So they need to be developed ethically and technically where the major problem arises in the field of technology. Due to the lack of technical work, one cannot have the problem be known to others in correct time. This application let nearby police to know their exact location and can help them in time. All the victim need to do is to click the button in the application so that the exact co-ordinates of their location including latitude and longitude values will be sent. The same information is also sent to their parents in the form of a text message. With such application the crime rate gets decreased and can also be helpful to government in achieving it. It doesn’t need any other member for police to inform. They can be the only one who can save themselves by just one click. In such a way, this application helps them to pass information and being saved from cruel. Also we have done multiple tests on different systems using which support Android with GPS on mobile and messaging system supportability.

Wireless Sensor Network (WSN) is a system with huge number of sensors connected to one another by placing them in a specific area. Different issues with WSN includes (but not limited to) the coverage, network lifetime and aggregation. The lifetime of a network can be improved by the clustering with the reduction of energy consumption. Clustering will group the related type of sensors into a single place with a head sensor node for message aggregation and transmission between other nodes and Base Station (BS). The cluster head (CH) consume more energy, when aggregating and transmitting the data. With the suitable identification of CH, there will be a reduction in the consumption of energy and improves the life of Wireless Sensor Network to be more. This paper modifies the meta-heuristic algorithms for improving the network lifetime by choosing appropriate cluster head and optimal path. K-Genetic Algorithm (K-GA) is proposed for efficient cluster head selection. Initially, the sensors are clustered using k-means clustering based on their location and Genetic Algorithm has been applied to detect the best cluster head. For secure optimal routing, Trust based Firefly (T-FA) path selection algorithm is used. Extensive simulations are conducted on various circumstances. The results obtained on the simulation indicates that the proposed K-GA helps in determining the optimized head of the cluster and T-FA discovers the optimal paths which enriches the life of the network by reducing end-to-end delay compared to other techniques.

In recent years everything is connected and passing through the internet, but Internet of Things (IOT), which will change all aspects of our lives and future. While the things are connected to the internet, they will generate the huge amount of information which has to be processed. The information that gathered from various IoT devices has to be recognized and organized according to the environments of their type. To recognize and organize the data gathered from different things, the important task to be played is making things passing through different Data Mining Techniques (DMT). In this article, we mainly focus on analysis of various Data Mining Techniques over the data that has been generated by the IOT Devices which are connected over the internet using DBSCAN Technique. And also performed review over different Data Mining Techniques for Data Analysis.