IDRandom-Forest: Advanced random forest for real-time intrusion detection

Abstract

In the last decade, with the increase in cyberattacks the privacy of network traffic has become a critical issue. Currently, simple network intrusion detection techniques are inefficient in terms of time complexity and are characterized by low detection accuracy and high false alarm rates, whereas techniques using complex algorithms such as recurrent neural network (RNN) and transformer-based deep learning, face challenges of high time complexity, large computational resource usage, and high latency rate in detecting intrusion in real-time traffic. To overcome these issues, we propose an advanced intrusion detection random forest “IDRandom-Forest” for real-time intrusion detection with reduced testing time and with higher accuracy. In this technique, an accuracy sliding window and feature weighting based on stratified feature sampling are introduced to determine the optimal sub-ensemble from the classical random forest. Experimental results demonstrated that the proposed hybrid classification system outperforms current state-of-the-art techniques in terms of accuracy and testing time.