M KALAISELVI

@drngpit.ac.in

Assistant Professor(SG),Dr. N.G.P. Institute of Technology, Coimbatore.
Dr.N.G.P institute of technology

M KALAISELVI


                  Download Compact PDF  
https://researchid.co/kalaiselvimayilsamy

RESEARCH, TEACHING, or OTHER INTERESTS

Computer Engineering, Engineering, Multidisciplinary, Computer Networks and Communications
3

Scopus Publications

Scopus Publications

  • Intrusion Detection and Security Challenges in 6G Networks Using Stochastic Graph Neural Networks
    Jeba Praba. J, M. Kalaiselvi, Sandeep Kumar Dasa, Neeru Malik, Rani Venkata Satya Praveen, Udayakumar N
    2025 International Conference on Information Implementation and Innovation in Technology I2itcon 2025, 2025
    While 5G is famous for its cloudification and micro-services-oriented design, intelligent network orchestration and management are key to the 6G era of networks. Therefore, the 6G paradigm that is being conceived relies heavily on AI, ML, and DL. Proactive threat identification, smart mitigation tactics, and assurance that 6G networks would be self-sustaining are essential for future end-to-end network automation. 6G communications will allow consumers to interact with online virtual worlds after 2030. For each separate analysis of the Network Intrusion dataset (CIC-IDS-2017), it chooses the optimal subset feature and lowers dimensionality. For the purpose of aggregation, the voting average approach is employed, and two classifiers—LSTM, GNN, RLGNN, and BiGRU—are transformed into DL algorithms. The current classification approach was surpassed by the suggested GNN method. According to the CIC-IDS-2017 Network Intrusion dataset, the accuracy rate was 96.54%.
  • Advanced Anomaly Detection in E-Commerce Fraud Prevention Using a Hybrid XGBoost-DNN Model
    A. Venu Gopal Reddy, M. Kalaiselvi, K. Kartheeban, D. R. Prince Williams, Rajanish Kumar Kaushal, Umanesan R
    2025 International Conference on Smart and Sustainable Technology Incsst 2025, 2025
    The swift growth of online shopping and electronic commerce has resulted in a substantial surge in electronic transactions, thus leading to a corresponding increase in fraud cases that incur billions of dollars in losses for firms each year. The escalating threat has rendered E-Commerce Fraud Prevention a vital domain of inquiry. The suggested work presents a novel fraud detection model consisting of three essential stages: Preprocessing, Feature Extraction, and Model Training. Data normalisation is implemented during preprocessing to guarantee quality, while the TF-IDF method is employed to extract pertinent features. Feature selection is performed with the XGBoost technique, succeeded by classification through a hybrid XGB-DNN model. The Deep Neural Network is optimised using the Adam optimiser to improve learning efficiency, and a softmax classifier is utilised for the final fraud classification. The experimental findings indicate that the suggested model substantially surpasses current shallow learning methods, attaining notable performance metrics: $\mathbf{9 9 . 3 9 \%}$ accuracy, $\mathbf{9 7 . 8 4 \%}$ precision, $\mathbf{9 5 . 4 8 \%}$ recall, $99.71 \% \mathrm{AUC}$, and $99.56 \% \mathrm{~F} 1$-score. The results validate the model’s efficacy in accurately identifying fraudulent transactions while minimising false positives. The study offers a comprehensive, data-driven methodology for enhancing E-Commerce Fraud Prevention, hence fostering safer and more secure digital commerce settings.
  • Anomaly Detection of Smart City in IoT Cyberattacks based on Hybrid A-BiLSTM-CRF Method
    Amit Jaykumar Chinchawade, M. Kalaiselvi, T. Sathish Kumar, Yogendra Singh, S. Srisathirapathy, G. Charles Babu
    International Conference on Sustainable Communication Networks and Application Icscna 2023 Proceedings, 2023
    In recent years, the new Internet of Things (IoT) paradigm has gained widespread adoption, and as a result, smart cities have evolved. “Smart cities” deploy their solutions in real time to enhance the convenience and happiness of urban dwellers. The network traffic of a smart city via IoT systems is rising dramatically, generating new cyber security issues, because these IoT devices are being connected to sensors that are directly connected to massive cloud servers. Researchers need to improve techniques for detecting compromised Internet of Things (IoT) devices in order to defend against these incursions. The proposed method employs the steps of preprocessing the data, selecting the features to use, and training the model. Data preprocessing in machine learning and statistics typically involves the use of normalization and standardization feature scaling techniques to convert the numerical values included in the dataset into the desired scales and ranges. Chi-square, Pearson's Correlation Coefficient, and mutual information are the three main metrics used in the feature selection process. A-BiLSTM-CRF is then used to train the models with the input features. When compared to the two most common alternatives, CRF and Bi-LSTM, the proposed method performs better. The proposed method was successful with a success rate of 97.32%.