The integration ofartificialintelligence (AI)in educationalsettings hasrevolutionized learning experiences while simultaneously raising critical privacy concerns. This chapter examines the complex landscape of AI-driven educational technologies and their implications for student data privacy. As educational institutions increasingly adopt AI-powered learning management systems, personalized tutoring platforms, and predictive analytics tools, the collection, processing, and storage of sensitive student information have become paramount concerns. This comprehensive analysis explores current privacy challenges, regulatory frameworks including FERPA and GDPR, technological solutions such as differential privacy and federated learning, and emerging best practices for protecting student privacy while maintaining AI system effectiveness. The chapter presents a balanced perspective on navigating the tension between educational innovation and privacy protection.

Ensuring secure and privacy-preserving authentication in web applications remains a critical challenge due to the limitations of conventional single-factor approaches, which are vulnerable to attacks and fail to account for dynamic user behaviors. Existing multi-factor authentication (MFA) methods often rely on static rules, exposing users to unnecessary friction or weak security under evolving threat conditions. To address these gaps, this study proposes PPAB-RL, a Privacy-Preserving Adaptive Biometric framework leveraging Reinforcement Learning for intelligent MFA selection. The proposed method integrates homomorphic encryption for secure fingerprint feature storage, contextual risk scoring based on device, behavioral, and geolocation deviations, and RL-driven adaptive MFA to dynamically select authentication pathways from password-only to multi-step biometric verification. Implementation is carried out using Python, with biometric processing performed on the SOCOFing dataset containing 6,000 fingerprint images, and blockchain-enabled logging for immutable and tamper-proof audit trails. Experimental results demonstrate that PPAB-RL achieves 96.8% authentication accuracy, surpassing traditional password-only (84.2%) and fingerprint-only (93.5%) methods, while maintaining low encrypted matching overhead and minimal user friction. Ablation studies confirm the essential contribution of each module, biometric preprocessing, encryption, risk analysis, and RL-based adaptation to overall system robustness. The RL policy converges rapidly, allowing real-time adaptation to changing user behaviors and threat contexts. Overall, the proposed PPAB-RL framework establishes a highly secure, intelligent, and scalable authentication paradigm, combining encrypted biometrics, dynamic risk assessment, and blockchain validation, offering an innovative approach that can inspire further research in next-generation privacy-sensitive authentication systems.

The LIC (Life Insurance Corporation) has many different types of policies with increasing complexity, and additionally, with the increase in demand for personalised digital services, there is an expectation of intelligent recommendation systems that can interpret user intent and provide relevant product suggestions that extend beyond the use of keyword matching. We developed a semantic retrieval-based LIC Policy Recommendation System using transformer embedding models to find the best matching LIC Policy for each user's needs. We compiled all LIC Policies into a single canonical text corpus and created a domain-specific dataset by combining this corpus with user query data. We evaluated many forms of embedding approaches (TF-IDF, transformer, etc.) and focused primarily on Sentence-BERT (MPNet) and an LIC domain term fine-tuned version. LIC Policies were indexed using FAISS and HNSW to allow for scale via efficient similarity searches across thousands of policies. The experimental results showed that the fine-tuned system consistently outperformed the base system in all major information retrieval metrics (Precision@1, Precision@2, Recall@5 and Mean Reciprocal Rank [MRR]). More relevant policies found in higher-ranking retrieves indicate a strong semantic understanding and alignment with user intent for this system. This research lays the groundwork for developing a production-level LIC recommendation system, demonstrating the benefits of utilizing both domain-adaptive embeddings and hybrid retrieval methods. Continuing enhancements could include providing richer fine-tuning data, BM25 Lexical Signals reranking, and offering real-time recommendations based upon user profiles.

Real-time surveillance systems require accurate and efficient object detection to ensure safety and situational awareness. Existing methods, such as YOLOv5 and Vision Transformer-based detectors, often struggle to reliably identify small, distant, or occluded objects while maintaining real-time inference, limiting their applicability in complex surveillance environments. To address these challenges, this study proposes PRISM, a hybrid Transformer–YOLOv8 framework that integrates fast local feature extraction with global contextual refinement. The method introduces two novel components: i) a Context-Aware Feed Forward Network (CA-FFN) within the Vision Transformer (ViT), which dynamically weights channel features to reduce redundancy and enhance global context modeling, and ii) Cross-Scale Attention Skip Connections (CSASC) for selective fusion of multi-scale YOLOv8 and ViT features, improving detection of small or occluded objects. The model is implemented in PyTorch and trained on a comprehensive surveillance dataset consisting of pedestrians, vehicles, bicycles, bags, and miscellaneous objects. Experimental evaluation demonstrates that PRISM achieves 96% accuracy, a significant improvement of ~4–5% over baseline methods, with robust performance across all object categories. Key performance indicators verify the reliability of the model to real-time usage, and the lightweight design makes it edge deployable. These findings imply that PRISM can be used to provide a speed-accuracy balance in a complex and dynamic setting, which is more efficient than the current methods. The study also notes the partial extensions, such as the incorporation of multi-sensors and continuous video streams to do time modeling as an extension, which will offer a good base to the next-generation intelligent surveillance systems.

Over the past few years there has been a tremendous increase in the worldwide prevalence of goiter and thyroid-based ailments, hitting people belonging to vast ages. Since the thyroid gland is instrumental in the regulation of most important metabolic processes it is pertinent that thyroid malfunctions be identified early and correctly. Nevertheless, although there are numerous available diagnostic algorithms that may be used, they either could only signal a narrow range of thyroid disorders or did not cover other problems like dataset imbalance in classes. This paper develops a diagnostic model that is robust and flexible, by incorporating the deep feature utilization performance of ResNet101 and the ensemble learning powerfulness of the AdaBoost classifier. The suggestion of ResNet101+ AdaBoost can be used successfully to detect ten classes of goiter and thyroid disorders in a single framework. With an enviable 96% accuracy, which when compared with base classifiers like Naive Bayes (88%) and Logistic Regression (91%).The ResNet101+AdaBoost model is highly reliable as a diagnostic tool because it overcomes data imbalance and identifies even intricate patterns in medical imaging. The system promises an application in global healthcare delivery because it detects related conditions such as thyroid and goiter conditions early and improves the outcome of treatment.

Federated learning enables collaborative model training across distributed healthcare institutions without centralizing sensitive patient data. However, existing approaches face challenges in balancing privacy protection and model utility. This paper proposes an adaptive differential privacy mechanism for federated learning in healthcare data aggregation that dynamically adjusts privacy budgets based on data sensitivity and model convergence. We introduce a privacy-utility trade-off optimizer that allocates privacy budgets adaptively across training rounds while maintaining rigorous privacy guarantees. Our experimental evaluation on the MIMIC-III and eICU datasets demonstrates that the proposed approach achieves 7.3% higher model accuracy compared to fixed differential privacy mechanisms while providing equivalent privacy protection with ϵ=2.5. The system reduces communication overhead by 31% through selective gradient aggregation and achieves convergence 23% faster than baseline federated learning approaches.

The scale of IoT and IIoT systems developing is growing rapidly since they deploy in critical infrastructure which also created major security issues such as data breaches, unauthorized access, and centralized model lack of trust issues. It is search of this study to formulate a block-chain-based IoT network that provides secure, trustful and corrupted-proof transmission of data. What is new in the approach is the combination of the use of machine learning-based real-time anomaly detection and the Ethereum-based smart contracts in the creation of a decentralized and intelligent security layer. In the offered framework, models XGBoost are trained using the Edge-IIoTset dataset with maximum accuracy of detection of 98%. Compared to traditional centralized and standalone ML-based solutions, the hybrid system is found to be much more efficient in both initial detection (precision), trust enforcement, and resilience. The findings ratify that the framework can be deployed in sensitive IoT/IIoT networks based not only on its ability to identify the source of cyber threats and curb them in real-time without necessitating any update, but also on its potential to increase the transparency and traceability of the network.

Nowadays with the emergence of Industry 4. 0, there is therefore the need to have sustainable manufacturing in order to increase operational efficiency as well as relieve the pressure on the environment. A potential weakness of conventional approaches is time-dependency while processing real-time data to prevent dysfunctions in the company's resource management, energy consumption control, and predictive maintenance. Thus, in this work it is proposed an innovative approach to address these problems with the use of Deep Reinforcement Learning (DRL) and Internet of Things (IoT) technology. Energy consumption, machine performance and environmental data are collected by IoT sensors in real-time and fed into a DRL model. This concept effectively eliminates the vices that are characteristic of conventional approaches in that it seeks to optimize production processes by cell design. The approach suggested above attains an unprecedented 98. 2% accuracy rate. This is in a bid to exemplify its capability in predicting these phenomenal and balance forecasts, to improve the overall performance and sustainability of manufacturing. Thereby, the method offers a robust answer to the acute questions arising in front-line manufacturing today and sets a brand-new reference for intelligent and sustainable behavior in the context of the fourth industrial revolution 4.0.

This research paper provides innovative approaches to support financial prediction, or it is a different kind of economic prediction that extends over collecting different economic information. Financial prediction is a concept that has been employed. The present study offers a unique approach to predicting finances by integrating many financial issues utilizing a cross-stitch hybrid approach. The method uses information from several financial databases, including market data, corporate reports, and macroeconomic indicators, to create a comprehensive dataset. Employing MinMax normalization the features are equally scaled to provide uniform input for the algorithm. The combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) systems form the basis of the framework. To understand the time-dependent nature of financial information, LSTM networks (long short-term memory) are utilized to record and simulate the temporal interactions and patterns. Concurrently, spatial features are extracted using CNNs; these components help identify patterns that are difficult to identify with conventional techniques. Better handling of risks, more optimal approaches to investing, and more informed decision-making are made possible by the enhanced forecasting potential that this method—which is described above—offers. Potential pilot studies will focus on innovative uses in financial decision-making and advancements in cross-stitching structure. This paper proposes a sophisticated approach that can help stakeholders, such as investors, analysts of data, and other financial intermediaries, traverse the complexities of financial markets.

In an increasingly diversified educational landscape, adaptable curriculum development is critical to meeting the varying requirements of students. Traditional curriculum design approaches frequently lack the flexibility to suit individual student variances, resulting in disengagement and poor learning outcomes. Current approaches are mostly based on static evaluations and generic material delivery, and do not take advantage of the wealth of data accessible regarding student achievement and participation. This work provides a unique way for optimizing adaptive curriculum development using Light Gradient Boosting (LightGBM), a machine learning method known for its effectiveness and predictive capacity. By using real-time data analytics, the suggested system enables personalized learning pathways that dynamically alter content based on student progress and preferences. The overall methodology includes collecting data from several educational sources, pre-processing to guarantee quality, and using LightGBM for predictive modelling. The adaptive curriculum's efficacy is evaluated using key measures such as pupil involvement, rate of retention, and academic performance. A series of case studies from various educational settings throughout the world are used to evaluate performance, comparing traditional curricula to an adaptive model constructed using LightGBM. Preliminary data show considerable gains in student outcomes, including greater engagement and achievement levels.

In the ever-evolving landscape of media, the demand for efficient and robust analysis of images and videos has intensified. Traditional methods often struggle to keep pace with the scale and complexity of media data. In response, this study introduces a novel approach that integrates Federated Learning (FL) and Bidirectional Long Short-Term Memory (BiLSTM) networks to enhance the analysis of images and videos in media applications. Federated Learning, a decentralized machine learning technique, enables collaborative model training across multiple edge devices without centralized data aggregation, thus addressing privacy concerns and data silo issues inherent in traditional approaches. By leveraging FL, The proposed framework facilitates the aggregation of insights from diverse sources while preserving data privacy. Furthermore, the integration of BiLSTM networks offers enhanced temporal modeling capabilities, allowing for the extraction of contextual information from sequential data such as video frames.Through experimentation on diverse media datasets, including images and videos, demonstrate the effectiveness of approach in tasks such as object recognition, scene understanding, and action recognition. The results showcase significant improvements in accuracy and efficiency compared to baseline methods, highlighting the potential of Federated Learning and BiLSTM integration for advancing image and video analysis in media applications.Overall, This study contributes to the ongoing efforts to innovate media analysis techniques by harnessing the power of decentralized learning and advanced sequential modeling, paving the way for more intelligent and privacy-preserving media analysis systems. This method achieves an accuracy of 97.5% and has been implemented in Python.

Optical Character Recognition (OCR) have an importance in the research based on image processing and recognizing the pattern. It serves as an automatic technique for identifying the various patterns in diverse applications. The recognition technique effectively explores the characters, images, and even the handwriting of an individual. Much research concerning OCR involves every deep learning, machine learning, and artificial intelligence algorithm. A segmentation-free approach has been introduced in this paper that combines with Hybrid Network Model (HNM) that works on collaborating convolutional neural network and recurrent neural network for minimizing the time in processing and improving the accuracy. In hybrid model we consider CNN in the input layer, the middle layer is LSTM and MLP is the layer generated as an output. The length of sequences in the input and output can vary it need not be specific they are managed by the encoder and decoders present in LSTM.

Nowadays Machine Learning is the most useful area for students and researchers for classification and prediction by using the machine learning algorithms. In this four kinds of algorithms are available. Those are supervised, unsupervised, semi-supervised and reinforcement algorithms. In this paper is compared the already existed KNN algorithm and a new Dynamic KNN algorithm with some case studies in order to show the accuracy. Dynamic KNN algorithm is an improvised version of KNN only. One of the significant factors that affect the accuracy of the model is the value of K. Fundamentally; the value of K is fixed for all data points. In this study, the value is calculated at the local level for all test points. This study performs a comparative study between suggested model and the traditional KNN model.

The best biometric information processes is a face recognition device, its applicability is simpler and its working range is broader than other methods like fingerprinting, iris scanning and signature. Face Detection is one of the kinds of bio-metric strategies that immediately apply to facial recognition by computerized devices through staring at the facial. It is a common feature used in bio analytics, digital cameras, and social labeling. Main applications of facial recognition algorithms that concentrate on recognition of face include environments, artifacts, and other parts of humans. Face-detection systems uses learning algorithms which are part of machine learning that can be used to identify subject faces inside big size pictures in order to function.

Two typical methods are used to choose a stock. The first is an essential analysis and the second is a study of technical aspects. Basic analysis and technical analysis, however, use totally distinct ways to collect stocks. Fundamental analysis and technical analysis may be used to evaluate whether or not an investment in the stock is lucrative and to predict future stock trends further. Basic and technical analysis to produce a comprehensive trading strategy may be blended. The aim of this document is to provide the top basic analysis and stock valuation methods for selecting stocks of actively traded stock portfolios used by everyday equity traders. Daily equity traders utilise mainly technical charts and other tools to identify patterns which may promote business perspective without evaluating the fundamental worth of a company to take trade choices. The examination of charts is designed to identify highly anticipated probability results in trades with precise price objectives. The objective of this technical paper is to highlight the importance of fundamental analysis in the everyday investment decisions of traders. The fundamental analysis focuses on important comparisons between the intrinsic value of an inventory and the current market price. If the intrinsic value of the stock exceeds the price marker, the acquisition of the stock by a fundamental investor/trader makes sense. This research supports the idea that equities traders should use both investing techniques to make more successful investment decisions.

There are two commonly held beliefs when it comes to picking stocks. Fundamental analysis is the initial step in making an investment decision, while technical analysis is the process used in making that decision. Investing involves using two separate analytical tools: fundamental analysis and technical analysis. Fundamental analysis and technical analysis can both tell whether an investment in a company is appealing or unattractive, and then go on to speculate on what the future trends of stocks will be. The combination of fundamental and technical research may provide a complete trading strategy. Artificial recurrent neural network (RNN) architecture, long short-term memory (LSTM) networks. The large-sequence-processing capabilities of LSTMs apply to many data sets. The vast quantity of data that is produced every day in the stock market is ideal for use in artificial intelligence applications. We want to use LSTM to build a financial market forecasting network that uses Technical and Fundamental analysis of businesses to predict the stock prices the next day. To do both kinds of analysis, the input data is pre-processed to contain necessary variables and then trained on LSTM& GRU. In this work we proposed Clustered Gradient Descent Adam optimizer usually perform better than models with Adam optimizer. The GRU model beats the LSTM model when it comes to overall accuracy.

In the world, every country depending on the economy. Economy depending upon the population activities. One of the most vital activities in these activities is selling, buying, and Investing money. So, Investment of the money on different aspects of the economy. So, Investment of the money common man to a rich person thinking about it. Above all the things are call share market. So many people are daily investing the daily money basis and long term basis also. Every significant business factor is risk analysis and trends of the share prices. In the share market, a massive amount of data generated day by day. So In this paper objective is to predict the low price and high price situations. This situation to predict using machine learning regression techniques is Linear, Lasso, Bayesian ridge, ridge, and Embaded Stack Regression algorithms. And predict the classes of high and low prices in time sequences and find out the performance attributes of the regression and compare them with the new embed stacking algorithm.

We often meet product development requirements which need to analyse, interpret or transform data from different business sources. Most manufacturing industries use their product manufacturing purposes for high-cost equipment. The impact of a failure cannot be afforded and production will be suspended. The machine is subject to preventive maintenance, often time-in-service inspections and repairs, to avoid such an enormous loss. The challenge of proper planning is increasing with the complexity of machinery: working together and effecting the lives of one another in a system with several components. Predictive maintenance is aimed at developing models that evaluate the risk of failure of a machine at any time and use these analytics to enhance the maintenance schedule. The achievement of predictive maintenance frameworks relies on various principal components: that the right data are available, that the problem is properly framed and predictions are evaluated appropriately. This paper will provide insights on how to choose the best modelling technique for the daily usage, service life of the machine and how to predict certain breakthrough points.

The recent growth in data collection and management processes have made the financial or transactional data availability for every modelling process such as forecasting. The financial forecasting replies on data modelling on the transactional financial data and which again relies on traditional data mining processes. The data mining process, in spite of multiple advantages such as modelling flexibilities, the sub-processes must be customized for fitting into the financial modelling. A good number of research attempts were made to predict the financial data points based on time series transactional data. However, most of the parallel research outcomes are criticised for higher complexity. Also, the outcome of the predictive or forecasting processes rely on the data cleaning methods, thus, the search for the less complex and composite data cleaning method is continuing as a focus of recent researches. The data cleaning method primarily concerns the outlier or noise detection and reduction, anomaly detection and reduction, missing value detection and reduction with the scope for dimensionality reductions. The parallel research outcomes for data cleaning is highly criticised for higher complexity. Thus, this work proposes a novel method for data cleaning with the help of composite data cleaning coefficient method. As a result, this work demonstrates nearly 100% accurate for any modelling technique with the proposed cleaning method, which is nearly 6% improvement to each modelling technique. In the course of establishing the novel method, this work also showcases the mapping of traditional data mining process into the financial forecasting models.

The Bitcoin market is filled with vast amounts of incomprehensible data. This work aims to provide a concise and simple representation of the data so that even the oblivious eye can make conclusions through the concepts of data mining and machine learning. Trading on Bitcoins is growing at a fast pace and is playing a vital role in data analysis and in the field of economics. On the other hand, Machine Learning is a part of data science that focuses on designing algorithm, such that the algorithm will understand the data provided by the user and predicts the value accordingly. This paper focuses on the market trends for popular bitcoins from various established sources like quandle, predicting for a future period of time using machine learning algorithm named Vector Auto Regression (VAR). This is highly advantageous as the investor, not only look at different stocks simultaneously but is also provided with highly accurate data, with an accuracy of 90%. We plan to take this to the next step by providing insight into the future market trends and allowing the investor to compare all the currencies in one platform and make his investment accordingly.