Optimality of Feature set for Intrusion Detection System

Intrusion detection systems (IDSs) became indispensable with the emerging requirement of security in computer network systems. Conventional detection techniques, like signature and rule-based intrusion detection, require regular human intervention or let the intrusion undetected. Fortunately, detection through Machine learning (ML) is free of such shortcomings. However, the selection of the most significant and predictive features is a challenge. The research community is quite active in the selection of the best subset of features in IDS. However, there is lacking a structured selection procedure and ordered list of features. We attempt to provide a more concrete list of features regarding their significance in predicting intrusion. We perform a survey and follow a structured methodology in features' selection out of the publicized dataset NSL-KDD. The features' selection procedure comprises five steps. The first three steps are dedicated to drop trivial features, while the last two steps are performed to identify the useful features. Model building is done using Support Vector Machine (SVM) for its wide acceptability in the IDS research community. The result comprises the ordered list of features. The features are sorted as per their predictive ability in classifying the malicious and benign network traffic.