Detecting Cyber Attacks: A Reinforcement Learning Based Intrusion Detection System

The network traffic is an ever increasing stream, sourcing from incrementing number of smart and network connected devices. The sources of the network traffic cannot be always trusted, resulting the requirement of a detection system that can flag malicious traffics as intruders. Such network traffic has numerous features and hence it is important to select only the relevant features of the traffic data, to develop an efficient intrusion detection system. Several intrusion detection studies focus on feature selection or reduction because some features are not correlated with the target variable. The presented research selects an optimal number of traffic data features and reduces the false positives with an unsupervised learning approach, during the application of intrusion detection system. The unsupervised algorithms selected for the implementation include K-Nearest Neighbors, Isolation Forest, and reinforcement learning, whereas reinforcement learning performs better than other semi-supervised learning methods.