MACHINE LEARNING-BASED NETWORK INTRUSION DETECTION USING RANDOM COMMITTEE ENSEMBLES: A MULTI-METRIC PERFORMANCE EVALUATION

Abstract

Network intrusion detection systems (NIDSs) are used as a tool to prevent the network from various attacks. Network intruders use the software's flaw to launch a number of assaults against computer network security, which results in the loss of key data. To this end, researchers in the recent past presented various intelligence-based models, each with its strengths and weaknesses. To design a proactive protective system, machine learning (ML) is widely used to monitor and respond to any cyber threats quickly. To keep the discussion of previous studies, this study proposes Random Committee (RC), an ML-based model for NIDS, which is a supervised ML technique and applied to labeled data. The Proposed model results are also compared to a number of Machine Learning models, which include Random Tree, Hoeffding Tree, Decision Stump, Decision Table, K-Nearest Neighbor, Naive Bayes, Boosting, and Bagging using the Waikato Environment for Knowledge Analysis (WEKA) tool. The assessments are made using multiple metrics, which are the Matthew correlation coefficient, false positive rate, true positive rate, accuracy, precision, and receiver operating characteristic area. The Kaggle and NSL-KDD datasets are used in both training and testing. The findings indicate that the proposed models are superior and have an accuracy of 99.9, and this forms a baseline for future studies. In turn, these results are the baseline for the researchers in terms of deciding and setting the priority of cyber-related properties in achieving a successful and best NIDS.