Entropy-Based Feature Selection for Network Intrusion Detection Systems

Abstract

A network intrusion detection system (NIDS) has a significant role in an industry or organization to protect their data. NIDS should be more reliable to manage huge traffic over the networks to detect the emerging attacks. In this chapter, novel entropy-based feature selection is proposed to select the important features of intrusion detection system. Feature selection reduces the computational time and improves detection rates. In entropy, within-class entropies and between-class entropies are computed for the various classes of intrusion in the KDD dataset. Based on computed entropy values, features are ranked and selected. Radial basis neural network (RBNN) is employed as a classifier. Performances of the proposed entropy-based feature selection algorithm are evaluated using the 10% dataset for training and two other datasets for testing. The proposed system shows significant improvement in the detection rate, reduces the false positive rate (FPR), and also reduces the computational time.