Cyber Threats and Network Intrusion Detection: A Comparative Analysis of Soft and Hard Voting Ensemble Methods

Abstract

More reliable techniques for detecting network intrusions are needed due to the growing supremacy of cyber attacks. Despite using signature-based techniques, traditional intrusion detection systems (IDS) are still not very successful against new threats. A thorough comparison of soft and hard voting ensemble techniques for binary network traffic classification is provided in this article. To create a powerful classifier system, the suggested approach combines five distinct supervised machine learning (ML) techniques. This study uses the UNSW-NB15 dataset, and correlation analysis is used extensively in feature selection to determine the optimal intrusion detection settings. According to experimental results, group approaches reach 99.73% accuracy utilizing the soft vote technique and surpass standalone classifiers. Excellent results from the classifier system include 99.22% precision, 99.71% recall. The classifier system has excellent performances such as 99.22% precision, 99.71% recall, and 99.46% F1 score, which indicates an extremely low false positive rate of 0.00421.