Intrusion Detection System Based on Machine and Deep Learning Models: A Comparative and Exhaustive Study

Network Intrusion Detection System plays a central role in detecting various security breaches and cyber attacks on a network. Literature suggests that machine learning techniques can successfully be used for intrusion detection, but there are many open challenges in the domain. In this paper, we performed multi-class classification for intrusion detection using different machine and deep learning techniques on four publicly available datasets. Our work focuses on evaluating the performance of the different intrusion detection models and achieving a better detection rate. Our experimental results show that hybrid CNN-LSTM and kNN models achieved an accuracy above 99% on KDDCup 99, CIC-IDS2017, and Bot-IoT datasets. These models also attain a detection rate of more than 0.9 for the DoS & DDoS attacks and an average FPR of less than 0.1 across all four datasets.