Computer Engineering, Computer Networks and Communications
38
Scopus Publications
Scopus Publications
AI-driven real-time anomaly detection in industrial cyber-physical systems using hybrid deep learning with SWaT and WADI benchmarks Smilarubavathy G, Nidhya R, Pavithra D, V B Thurai Raaj, Keerthana S M Franklin Open, 2026 Cyber-Physical Systems (CPSs) are increasingly exposed to sophisticated cyberattacks, necessitating the development of robust, real-time anomaly detection solutions to maintain system reliability and operational continuity. This study proposes a hybrid deep learning approach that combines CNN, BiLSTM, and an attention mechanism to detect the anomalies in an industrial CPS environments. This architecture uses (a) CNNs to extract spatial features from input data; (b) BiLSTM networks to model temporal sequences; and (c) an attention mechanism included to emphasise the significant trends and enhance the interpretability. The model uses the sliding window inference technique to enable the real-time usage. It also includes an adaptive thresholding method based on statistical estimation. The system is tested on two popular CPSs benchmark datasets: Secure Water Treatment (SWaT) and Water Distribution (WADI), which include 36 labelled attack scenarios and multiple stealthy attack patterns, respectively. Experimental results shows that the proposed model achieves high accuracy (98.7% on SWaT and 97.4% on WADI) and F1-scores above 97.5%, with the inference latency below 50 ms per prediction. Compared with baseline CNN, LSTM, and SVM detectors, the hybrid approach improves precision under noisy and imbalanced conditions, demonstrating superiority in real-time CPS anomaly detection. An ablation study verifies the individual impact of each component in the model. Furthermore, this study presents a theoretical detection-delay bound, and confirming the framework’s suitability for real-time CPS monitoring.
Amazon Bedrock Medical Bot: Revolutionizing Patient Care D. Pavithra, Divya M Proceedings of 8th International Conference on Inventive Computation Technologies Icict 2025, 2025 This research presents an AI-driven medical chatbot system leveraging AWS services to provide intelligent responses for minor medical symptoms, medical record analysis, and prescription interpretation. The architecture integrates multiple AWS components to ensure seamless text extraction, data processing, and contextual understanding. Users interact via Slack, submitting documents, PDFs, or images containing medical queries. Amazon Lex processes these inputs, directing them to Amazon S3 for secure storage. Amazon Textract extracts relevant text, which is then analysed by Amazon Comprehend Medical to identify key entities such as symptoms, medications, and diagnoses. Amazon Bedrock further enhances this workflow by employing advanced AI models to generate context-aware medical insights, offering users precise and reliable guidance. By automating text extraction, entity recognition, and medical data interpretation, this system enhances decision-making efficiency and accessibility in healthcare, making expert-level analysis more readily available to users.
Personal Finance Tracker with Spending Behavior Analysis D. Pavithra, Dhivyaprabha N, Shreenithi K, Theshika C C Proceedings of 5th International Conference on Evolutionary Computing and Mobile Sustainable Networks Icecmsn 2025, 2025 Personal Finance Tracker framework that incorporates machine learning and advanced data analytics to provide predictive insights and behavioral insights into individual spending behavior. Current financial tools provide only backdate financial summaries to clients. The proposed framework emphasizes anticipatory and behavioral financial management. Users can introduce transactions manually. The system will then employ time series forecasting models such as Hybrid Autoencoder to help indicate forecasts of possible changes in cash flow. A significance of this work, is an approach utilizing unsupervised learning techniques, specifically HDBSCAN clustering, to group users' spending behavior and reveal latent behavioral profiles. These insights allow the framework to deliver context-aware alerts and personalized recommendations for possible savings. The platform also includes gamified monthly savings challenges, and visual analytics in real time to promote active user engagement, where tracking finances becomes fun and educational. Utilizing predictive modeling, behavioral measures, and human-centered design impacts financial literacy, engagement, and sustainable financial responsibility. The experimental evaluation and user feedback suggests that this hybrid approach can potentially become a basis for future generations of intelligent personal finance applications.
A Dual-Authentication Router-Assisted Attendance System D. Pavithra, Tarun Prasanth S, Nishore N, Kabinesh S Proceedings of 6th International Conference on Iot Based Control Networks and Intelligent Systems Icicnis 2025, 2025 Using a dual-authentication strategy that combines Wi-Fi-based validation and One-Time Password (OTP) verification, the DA-Router Framework presents an intelligent attendance management system that improves accuracy and stops proxy marking. Time inefficiency, human error, and spoofing are some of the problems with traditional attendance techniques, such as manual and biometric systems. Only students who are physically connected to the classroom’s assigned network (by SSID/BSSID/MAC verification) are able to mark attendance thanks to the suggested framework’s use of router-based device authentication. Simultaneously, identity validation is strengthened by an OTP created by the instructor and validated via a mobile or web interface. Session logs, device information, and attendance analytics are all stored in a centralized database that is updated in real time. A fallback QR mechanism guarantees continuity in the event of Wi-Fi outages. The system offers an administrator dashboard for monitoring attendance statistics, student performance, and connectivity history. The DA-Router Framework advances the automation and integrity of attendance management by enabling a dependable, scalable, and impenetrable solution for contemporary educational institutions by fusing network-layer verification with OTP-based security.
Automated Weed Detection in Crop Fields Using Convolutional Neural Networks: A Deep Learning Approach for Smart Farming Nidhya R, Pavithra D, Smilarubavathy G, Mythrayee D Data and Metadata, 2025 Deep learning is a part of modern machine learning that includes deep belief networks, deep neural networks, and recurrent neural networks. Computer vision, audio processing, and language comprehension are the most important sectors of deep learning. In many instances, these applications exceed human performance. In smart agriculture, deep learning gives novel ideas for increasing productivity and efficiency. Weed identification is an important application in crop areas that improves farming. This technology improves crop yields by identifying weeds. Also, it reduces resource wastage in agricultural practices. This paper presents a Convolutional Neural Network (CNN) model specifically designed to accurately identify and classify weeds using images of crop fields, augmented by the ImageNet dataset for enhanced feature extraction and model training. The model identifies essential characteristics, such as dimensions, form, spectral reflectance, and texture, to distinguish between crops and weeds. Unlike existing systems, our CNN-based approach achieves a high accuracy of 98%. This improvement enhances weed identification efficiency and reduces pesticide usage, therefore it minimising environmental impact.
Detection of IoT Attacks Using Hybrid RNN-DBN Model D. Pavithra, R. Bharathraj, P. Poovizhi, K. Libitharan, V. Nivetha Generative Artificial Intelligence Concepts and Applications, 2025 Cyberattacks are constantly evolving, becoming more complex and harder to detect. Hackers develop new exploits, malware, and phishing techniques faster than traditional security measures can adapt. Securing devices, against cyber threats is of importance in the context of the Internet of Things where all things are interconnected. The suggested framework presents a method, for detecting intrusions in networks by utilizing a model that combines recurrent neural networks (RNNs) and deep belief networks (DBNs). The RNNs excel at understanding patterns in sequential data, like network traffic flow in an Internet of Things environment. The DBNs are adept at extracting key features from data. By utilizing these methods, the goal is to create a system that can effectively identify and classify attacks. To assess its effectiveness this combined approach was applied to the UNSW NB15 dataset, which consists of both network traffic and malicious activity from a network. The outcomes show encouraging performance as the model accurately identified nine types of attacks such, as Backdoors, Denial of Service (DoS) Analysis, Exploits, Generic Threats, Reconnaissance activities, Fuzzers, for Abnormal Activity Shellcode Exploits, and Worms.
Machine Learning in Healthcare: Unlocking Precision Diagnosis and Continuous Monitoring Through Voice Analysis G. Smilarubavathy, S. M. Keerthana, R. Nidhya, Thanga Priscilla, D. Pavithra Smart Factories for Industry 5 0 Transformation, 2025 This chapter explores the emerging use of machine learning practices for voice analysis in modern healthcare. Precisely, it inspects the abilities and possible impact of applying artificial intelligence to extract medically applicable data from voice recordings. Machine learning models and Voice-based biomarkers are poised to transform healthcare by providing non-invasive, continuous remote patient monitoring and early diagnosis for an extensive range of diseases. However, significant challenges remain around rigorous validation, clinical integration, comparative assessment, ethics, and privacy of the patient. This chapter offers a complete overview of the advancement in this revolutionize field. It investigates the significance of vocal biomarker analysis with the help of machine learning methods for enhancing disease screening and disease diagnosis. An experimental study is obtainable by applying convolutional neural networks to examine the vocal feature data extracted from patient's audio samples with various conditions. Results demonstrate over 90% accuracy, categorizing six diseases which embrace depression, hypertension, and Parkinson's solely from voice. These outcomes highlight the capacity of vocal biomarkers and deep learning to permit non-invasive, low-cost, and early disease detection. With additional validation, voice-based AI could offer transformative abilities in telehealth, home monitoring, and precision medicine. This chapter provides valuable perceptions in both the remarkable promise and crucial considerations of executing machine learning's stethoscope in today's medicinal world.
YouTube Comment Analysis Using LSTM Model D. Pavithra, P. Poovizhi, G. Rokeshkumar, T. Bharathvaj, M. Mageshkumar Generative Artificial Intelligence Concepts and Applications, 2025 Addressing the ever-expanding landscape of online content consumption, this study introduces a groundbreaking approach to real-time sentiment analysis and comment categorization on platforms such as YouTube. Given the remarkable rate of video uploads, with a new video posted every minute globally, and an impressive daily consumption of content reaching one billion hours, the manual analysis of user-generated content has become impractical. The proposed system leverages long short-term memory (LSTM) recurrent neural networks (RNNs), specifically a stack of three LSTM layers, for effective sentiment analysis, overcoming challenges like vanishing and exploding gradients. In addition to sentiment analysis, the system employs an incremental updation approach for swift comment analysis and storage in the database, ensuring real-time analytics. Taking a step beyond traditional sentiment analysis, the system introduces a novel feature – comment categorization. Comments are not only analyzed for sentiment but are also categorized and stored, providing content creators with a segregated view of user feedback. This innovative approach enables content creators to enhance their videos based on specific comment categories, fostering an environment for improved content creation and audience engagement. In summary, this system represents a significant stride towards revolutionizing online content analysis, offering real-time insights and empowering content creators for a more dynamic and interactive content creation process.
Preface R. Nidhya, D. Pavithra, Manish Kumar, A. Dinesh Kumar, S. Balamurugan Generative Artificial Intelligence Concepts and Applications, 2025
Multimodal Biometric Fusion System Poovizhi P, Pavithra D, Vijaya Kumar T, Nandhakumar M, Suryadharshini G, Sheljin S G 2nd International Conference on Emerging Research in Computational Science Icercs 2024, 2024
Identification of Coal Gangue in Coal using YOLO V7 Pavithra D, Poovizhi P, Nikil Vignesh R, Vigneshwaran R, Sowrow A 7th International Conference on Trends in Electronics and Informatics Icoei 2023 Proceedings, 2023
