In the past decade with the advancement of networking, the network traffic size and complexity have grown. Alongside this, malicious activities are also increasing. One of the most popular methods for finding malicious activity in any network is to analyze its traffic using an intrusion detection system. There are many different approaches to designing intrusion detection systems, but the machine learning technique is the most effective. The performance of a machine learning-based intrusion detection system was negatively impacted by the redundant and unnecessary information that was discovered in network traffic. To address such issues, the detection system can use feature selection alongside effective classifiers. In this paper, we implement a hybrid Feature selection where entropy-based infinite feature selection in the first stage and Eigenvector centrality, and ranking feature selection algorithms in the second stage to optimally reduce the size of the network traffic dataset. After that three machine learning classifiers i.e., ANN, KNN, and DT are used to classify network traffic as normal or attack. Various metrics, including accuracy, recall, precision, and f1-score, are utilized to validate and evaluate the performance of the suggested feature selection and IDS model on the NSLKDD dataset.

Feature selection (FS), Machine learning (ML), NSLKDD dataset, intrusion detection systems (IDS).