OPTIMIZATION APPROACHES BASED RELEVANT SPACE SELECTION FOR EFFICIENT INTRUSION DETECTION

Abstract

Intrusion Detection System (IDS) aims to protect a network from a variety of attacks threatening their integrity, confidentiality and availability. Most of the machine learning based IDS used different algorithms for feature selection and sample selection. It may lead to complexity even when those algorithms are used in a consecutive manner. In this paper, Uniform Design Parameterized Particle Swarm Optimization (UDPPSO), Kullback-Leibler Divergence Controlled Firefly (KLDCFF) and Hybridized KLDCFF with Reverse solution based Improved Teaching Learning Optimization (H-KLDCFF-RSITL) optimization algorithms are proposed to improve the performance of classifier for IDS while reducing the complexity in feature selection and sample selection. It can be achieved by simultaneously selecting the most relevant features and samples. In UDPPSO, uniform parameters are designed to resolve the randomness and time problem of Particle Swarm Optimization (PSO). However, the UDPPSO has low convergence problem and it is solved by KLDCFF in which the firefly technique is used where the fireflies are moved based on the Kullback-Leibler Divergence (KLD) divergence measure. The search space of fireflies is high in KLDCFF. An H-KLDCFF-RSITL is proposed to reduce the search space by using a teaching-learning technique.  Thus the proposed UDPPSO, KLDCFF and H-KLDCFF-RSITL selected the features and samples simultaneously and it is given as input to Replicator Neural Networks (RNN) classifier which classifies the intrusions effectively.