Feature selection and intrusion classification in NSL-KDD cup 99 dataset employing SVMs

TLDR: A new approach consists with merging of feature selection and classification for multiple class NSL-KDD cup 99 intrusion detection dataset employing support vector machine (SVM) to improve the competence of intrusion classification with a significantly reduced set of input features from the training data.

Abstract: Intrusion is the violation of information security policy by malicious activities. Intrusion detection (ID) is a series of actions for detecting and recognising suspicious actions that make the expedient acceptance of standards of confidentiality, quality, consistency, and availability of a computer based network system. In this paper, we present a new approach consists with merging of feature selection and classification for multiple class NSL-KDD cup 99 intrusion detection dataset employing support vector machine (SVM). The objective is to improve the competence of intrusion classification with a significantly reduced set of input features from the training data. In supervised learning, feature selection is the process of selecting the important input training features and removing the irrelevant input training features, with the objective of obtaining a feature subset that produces higher classification accuracy. In the experiment, we have applied SVM classifier on several input feature subsets of training dataset of NSL-KDD cup 99 dataset. The experimental results obtained showed the proposed method successfully bring 91% classification accuracy using only three features and 99% classification accuracy using 36 features, while all 41 training features achieved 99% classification accuracy.