The network traffic in the intrusion detection system (IDS) has unpredictable behaviour due to the high computational power. The complexity of the system increases; thus, it is required to investigate the enormous number of features. However, the features that are inappropriate and (or) have some noisy data severely affect the performance of the IDSs. In this study, we have performed feature selection (FS) through a random forest algorithm for reducing irrelevant attributes. It makes the underlying task of intrusion detection effective and efficient. Later, a comparative study is carried through applying different classifiers, e.g., k Nearest Neighbour ( k -NN), Support Vector Machine (SVM), Logistic Regression (LR), decision tree (DT) and Naive Bayes (NB) for measuring the different IDS metrics. The particle swarm optimization (PSO) algorithm was applied on the selective features of the NSL-KDD dataset, which cut down the false alarm rate and enhanced the detection rate and the accuracy of the IDS as compared with the mentioned state-of-the-art classifiers. This study includes the accuracy, precision, false-positive rate and the detection rate as performance metrics for the IDSs. The experimental results show low computational complexity, 99.32% efficiency and 99.26% detection rate on the selected features (=10) out of a complete set (= 41).
展开▼
