Dinesh Ezhilmurugu

Data Scientist with expertise in Machine Learning and Cloud Computing, skilled at transforming large datasets into actionable insights through advanced predictive modeling, automation, and scalable CI/CD pipelines. Proven success in customer segmentation and targeted analytics to drive strategic business decisions. Adept at leading cross-functional collaborations to solve complex, real-world challenges with data-driven solutions.

1) IoT Devices Intrusion Detection: Evaluating Feature Extraction Techniques using Machine Learning

• Developed a machine learning-based solution to detect system process attacks in IoT devices, enhancing security in IoT environments.
• Processed raw data from multiple sources, implementing undersampling techniques to balance the dataset for improved model training.
• Applied advanced feature extraction methods, including Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and polynomial feature generation, to optimize model performance.
• Evaluated multiple machine learning models—Random Forest, XGBoost, and Long Short-Term Memory (LSTM)—to identify the most accurate and robust approach for attack detection.
• Achieved high accuracy (95%) using LDA with XGBoost and Random Forest, outperforming other models in precision and recall metrics.
• Recommended the combination of LDA with XGBoost for efficient, high-accuracy detection of system process attacks in IoT environments.

2) Software Vulnerability Detection: Exploring Advanced Semi-Supervised Techniques

• Focused on addressing class imbalance in software vulnerability detection using advanced semi-supervised learning techniques, including Bi-LSTM, Transductive-SVM, and Graph Label Propagation.
• Implemented semi-supervised learning approaches to improve detection of rare vulnerabilities, where vulnerable instances were significantly outnumbered by non-vulnerable ones.
• Balanced the dataset through downsampling to ensure equal representation of vulnerable and non-vulnerable instances, enhancing model accuracy.
• Evaluated and compared model performance, with Graph Label Propagation emerging as the top performer, achieving 94.23% accuracy and high precision/recall for both classes.
• Recommended Graph Label Propagation as the optimal approach for vulnerability detection, providing a reliable solution for identifying critical vulnerabilities in software code.