SPECIFICATION-BASED INTRUSION DETECTION SYSTEM FOR 802.11 NETWORKS USING INCREMENTAL DECISION TREE CLASSIFIER

Abstract

The objective of this thesis is to present a specification-based IDS technique, named as Normalized information gain and Tie Breaking Threshold-based Decision Tree (N-TBTDT) that utilizes various data mining techniques to improve the IDS performance significantly. The main contributions of the proposed IDS are feature reduction using normalized information gain, the chaotic Particle Swarm Optimization (PSO) for feature extraction and improved Very Fast Decision Tree (VFDT) for intrusion classification. The bias compensation factor-based tie-breaking threshold promises the efficient decision tree construction, rather than the random selection of tie-breaking threshold. This shows a significant improvement in the detection accuracy of N-TBTDT. To evaluate the performance of N-TBTDT, two different scenarios are created. Firstly, the training dataset size is varied, and secondly, the number of attacks is varied from low to high. The N-TBTDT exploits different performance metrics such as detection accuracy, false positive rate, precision, F-Score, and classification accuracy.