An efficiency optimization for network intrusion detection system

With the enormous rise in the usage of computer networks, the necessity for safeguarding these networks is also increased. Network intrusion detection systems (NIDS) are designed to monitor and inspect the activities in a network. NIDS mainly depends on the features of the input network data as these features give information on the behaviour nature of the network traffic. The irrelevant and redundant network features negatively affect the efficacy and quality of NIDS, particularly its classification accuracy, detection time and processing complexity. In this paper, several feature selection techniques are applied to optimize the efficiency of NIDS. The categories of the applied feature selection techniques are the filter, wrapper and hybrid. Support vector machine (SVM) is employed as the detection model to classify the network connections behaviour into normal and abnormal traffic. NIDS is trained and tested on the benchmark NSL-KDD dataset. The performance of the applied feature selection techniques is compared with each other and the results are discussed. Evaluation results demonstrated the superiority of the wrapper techniques in providing the highest classification accuracy with the lowest detection time and false alarms of the NIDS.