Dr. S. Ashwini has completed her Bachelors in Information Technology in Sri Sairam Engineering college, Chennai. She has completed her Masters in Computer Science and Engineering in Rrase college of engineering. She has more than 7 years of teaching experience in various engineering colleges. Her areas of interest include Machine Learning, Deep Learning, IOT, Cyber Security. She has coordinated many international Conferences. She is an active participant in accreditation process and in teaching learning process. She has published more than 18 research papers in various reputed and referred journals and attended conferences. She has guided 15+ and guiding students in both UG and PG level.
RESEARCH INTERESTS
Dr. S. Ashwini has completed her Bachelors in Information Technology in Sri Sairam Engineering college, Chennai. She has completed her Masters in Computer Science and Engineering in Rrase college of engineering. She has more than 7 years of teaching experience in various engineering colleges.
34
Scopus Publications
Scopus Publications
An Efficient Deep Learning Approach for Video Content Authenticity Verification Ashwini S, Nikitha K S, Savitha S K Proceedings of the 4th International Conference on Intelligent Data Communication Technologies and Internet of Things Idciot 2026, 2026 The increasing sophistication of Deepfake generation methods presents significant challenges for distinguishing authentic videos from synthetically manipulated content, raising critical concerns across social, security, and legal spheres. To address this, a robust deepfake detection framework is proposed that integrates frame-level extraction, Multi-task Cascaded Convolutional Neural Networks (MTCNN) for accurate facial localization, and data augmentation to enhance generalization and reduce overfitting. Leveraging transfer learning, the system employs a fine-tuned VGG16 Convolutional Neural Network (CNN) for high-level spatial feature extraction. Trained and evaluated on the extensive Deepfake Detection Challenge (DFDC) dataset, the model achieves an accuracy of 97.24%, effectively distinguishing between genuine and forged content even under challenging conditions such as low resolution and compression artifacts. By processing videos frame-by-frame and applying a majority-voting mechanism for final classification, the framework demonstrates strong robustness and adaptability across diverse video qualities and durations. These results underscore the system's potential for real-world deployment in applications such as digital forensics, social media verification, and cybersecurity.
Fuzzy System for Environmental Monitoring S. Ashwini, R. Dhwarithaa, R. Nithya PARANTHAMAN, T. Preethiya, G. Ramya, G. Abinaya Blockchain and the Water Supply Chain Opportunities Challenges and Innovations, 2025 This chapter discusses environmental monitoring using a fuzzy system, with its signature ability to model even fuzzy relationships among environmental variables. It details fuzzy inference system's architecture, including fuzzification, rule-based reasoning and defuzzification. The chapter presents case studies including various fuzzy logic applications in air quality monitoring, water pollution assessment and climate change analysis, demonstrating its effectiveness in synthesizing multisource data to yield actionable insights. It analyzes merging fuzzy systems with artificial intelligence (AI) techniques such as machine learning (ML) and neural networks to enhance prediction accuracy and real-time environmental decision-making. ML techniques such as random forest and support vector machines have been widely used to improve fuzzy rule optimization. Techniques of optimized fuzzy rule reduction help to minimize computing complexity. By integrating IoT, edge computing and explainable AI techniques, the next generation of fuzzy systems will drive smarter, more sustainable decision-making in environmental science.
Future Research Directions for Blockchain in Metaverse Healthcare G. Ramya, T. Preethiya, R. Nithya Paranthaman, S. Ashwini, R. Dhwarithaa, G. Abinaya Blockchain Based Healthcare Management in the Metaverse, 2025 The inclusion of Blockchain in the Metaverse delivers opportunities for healthcare, contributes to the data security, maintains transparency and decentralization of data. The current Healthcare systems revolve around the immersive data and digitalization of data putting the metaverse as a frontier. This spawned telemedicine, virtual doctor consultations, training people and patient engagement. The integration of Blockchain with the Metaverse healthcare aims to enhance the privacy, security and reliability of data. The key applications of it are patient record management, providing access to valid credentials and decentralized clinical assessments. This chapter tries to provide a comprehensive overview of Blockchain's transformative potential in Metaverse healthcare, highlighting key areas for future research and development. This chapter aims to address several future research directions such as interoperability of devices/data for seamless services, need for privacy-preserving of patient's data, regulatory compliance, ownership of data and other ethical considerations.
Enhanced cardiovascular disease prediction: AMWOA-based feature selection and PyramidConvFormer-VAE fusion approach P. Nancy, M. Rajkumar, S. Ashwini, J. Jegan Amarnath Computer Methods in Biomechanics and Biomedical Engineering, 2025 Cardiovascular disease remains a major global cause of death. To address challenges of high dimensionality and data imbalance in heart disease prediction, this study proposes a novel framework integrating feature optimization and classification. An Adaptive Mutated Walrus Optimization Algorithm (AMWOA) effectively reduces feature dimensions, mitigating overfitting and reducing execution time. For classification, a PyramidConvFormer-Variational Autoencoder (VAE) model integrates CNN and transformer layers to extract local-global patterns. Final classification is performed via fully connected layers with softmax activation. Evaluated on the Cleveland dataset using five-fold cross-validation, the proposed method achieves 98.12% accuracy and 98.91% precision, outperforming existing prediction frameworks.
Advancements in Multi-Agent Large Language Model Systems for Next- Generation AI: Multi- Agent LLMs in Healthcare and Diagnostics Abinaya Gopalakrishnan, G. Ramya, T. Preethiya, R. Nithya Paranthaman, S. Ashwini, R. Dhwarithaa Advancements in Multi Agent Large Language Model Systems for Next Generation AI, 2025 Large Language Models (LLMs) have enabled AI research. Their easier design of new ways to handle challenges across a wide range of applications has increased this discipline's influence. Multi-agent LLM systems in diagnostics and healthcare can revolutionize clinical decision-making, precision medicine, and patient care. This chapter examines multi-agent LLMs' medical concepts, designs, and applications.Multi-agent systems can scale, modularize, and specialize and integrate several medical specializations and contextual knowledge. This chapters covers the technical implementation of these systems, including advanced Large Language Models, quality control measures, guardrails, self-reflection, integration with EHRs, and explainable AI for decision transparency. We discuss possible benefits with future directions, like integrating IoT devices and creating advanced natural language interfaces.
Privacy-Preserving Approach to Email Spam Detection Using Federated Learning Thejaswini R., Ashwini S., Chandramouli H., Sharmila Shanthi Sequeira, Deepak H A, Kiran Puttegowda 2nd International Conference on Intelligent Algorithms for Computational Intelligence Systems Iacis 2025, 2025 Email spam detection remains a critical task in maintaining secure digital communication. While machine learning has significantly improved spam filtering accuracy, conventional approaches often rely on centralized data collection, raising serious privacy concerns. In this paper, we propose a privacy-preserving email spam detection framework using Federated Learning (FL), which enables collaborative model training without sharing raw user data. Proposed evaluation on the Enron dataset shows that Decision Tree and Random Forest classifiers deliver the best performance, with the Decision Tree achieving an F1 Score of 0.900 and an AUC of 0.918, while the Random Forest attained an F1 Score of 0.8785 and an AUC of 0.8656 in the centralized setting. Under the federated learning setting, the same model maintained strong performance with an F1 Score of 0.822 and AUC of 0.861, demonstrating that effective spam filtering can be achieved without compromising user privacy. These results highlight the potential of FL in deploying secure and scalable spam detection systems for privacy-sensitive environments.
