Performance of Feature Reduction Techniques for Efficient Intrusion Detection System

Abstract

In Cybersecurity, Intrusion detection system is a tool that is used to identify the abnormal activity on the network in a timely manner. This is done by inspecting and analyzing the features of network packets to find any anomaly in them. In high dimensional traffic, the intrusion detection system needs to inspect and analyze all the features of a network packet to make the decision, which is an overhead. Feature reduction techniques are used to map the existing feature set to a new but less feature space to make the intrusion detection system more efficient. In this work we compared the performance of three commonly used feature reduction techniques, viz. Principle Component Analysis, Independent Component Analysis and Linear Discriminant Analysis. We used k-Nearest Neighbor and Naïve Bayes algorithm as classifiers. It was observed that Linear Discriminant Analysis resulted in improved accuracy of 95-100% and reduced false alarms between 0-0.5%.