Development of an efficient classifier using proposed sensitivity-based feature selection technique for intrusion detection system

Intrusion detection system protects an individual computer or network computer from suspicious data and protects the system from unauthorized access. In this paper, we propose a feature selection technique (FST) known as sensitivity based feature selection technique (SBFST) which selects relevant features from intrusion data based on the value of sensitivity. We compare various existing FSTs with the proposed SBFST from three different categories of NSL-KDD data set. Experimental results reveal that C4.5 with SBFST performs better than other existing FSTs and produce a high accuracy of 99.68% with 11 features and 99.95% accuracy with nine features for the multiclass and binary class problems respectively. It has also produced 99.64% accuracy for both multiclass and binary class problems respectively with six and seven features. The performance of proposed SBFST is also verified using the intersection of features, segment by segment with other FSTs and found to be better.