Intraclass and interclass correlation coefficient-based feature selection in NIDS dataset

摘要

Offline processing of network intrusion detection system (NIDS) dataset helps to mitigate network attacks. KDD CUP'99 was created using known network attacks in 1999. The changes in networking trend and advent of new attacks have raised the demand for new NIDS datasets. Recent test beds such as UNB-ISCX, SSENet, and ITD-UTM help in the evaluation and deployment of new network security algorithms. NIDS dataset generated from these test beds is used by the traditional feature selection methods, which focus on the whole dataset for the reduction of the number of features. Later, instead of whole dataset, selective dataset is used to choose the best feature set for a given attack. In this paper, the application of intraclass correlation coefficient and interclass correlation coefficient is proposed to achieve efficient target class-specific feature subset. It is envisioned that such a feature subset would help in countering an attack type. Experiments were conducted on classification algorithms to validate on four datasets the effectiveness of the feature subsets on pairwise target class with and without applying our proposed feature selection algorithm. It is found that the detection rate increased and execution time and false alarms decreased considerably after applying the proposed feature selection algorithm. Copyright (C) 2015 John Wiley & Sons, Ltd.