RESEARCH

Research Experience

• Graph Coloring Heuristic: A Delayed but Color Efficient Approach 

In this work, we developed a novel graph coloring heuristic combining constraints on normal and incident degree. Experiments on real and synthetic graphs showed improved color efficiency compared to traditional methods, albeit with longer runtimes. 

• Hash Based AS Traceback Approach against DoS Attack 

Introduced an Autonomous System (AS) traceback mechanism based on probabilistic packet marking, which allows the victim to   trace the attack-originating AS. Traceback on the AS level has several advantages containing a reduced number of routers involvement  for packet marking as well as the required number of packets to infer the forward path. We utilize the IP packet header to implement our packet marking methodology. Our results show that a victim site can trace the attack path with 33.25 packets on average. Additionally, we provide an encoding method to significantly reduce the false cases in path reconstruction.

• Deep Learning-Based Intrusion Detection Model: 

Developed a deep learning-based model tailored for network intrusion detection, achieving an impressive accuracy of 97.58% on benchmark datasets. This work, which has been published in a high-impact journal, underscores my commitment to advancing cybersecurity solutions through the innovative application of deep learning technologies. 

• "DSVDD-CAE" Model for Anomaly Detection

Conceptualized and designed the “DSVDD-CAE” model integrating Contractive Autoencoder with Deep Support Vector Data Description. This model achieved over 99% accuracy in detecting anomalies in IoT networks, demonstrating my machine learning expertise.

• Ensemble Model for Network Intrusion Detection

Developed an ensemble model combining Contractive Autoencoders and K-Means Clustering for intrusion detection. When tested on the NSL-KDD dataset, this model achieved a 92% F1 score, outperforming existing methods. I also conducted comprehensive analysis of CAE hyperparameters to optimize anomaly detection capabilities.