Ensembling tree-based classifiers for improving the accuracy of cyber attack detection

Abstract

Nowadays, the new generation of technology has completely relied on the network-based services. So the wide use of the Internet gives an opportunity to cyber attackers to target the systems which process and save vital information and disrupt their functionality. According to this, the need for finding a way to prevent these attacks and make computer systems more secured is essential, and cyber security turns to a fundamental concern for researchers. A well-known technology in detecting unusual access to the network is Intrusion Detection Systems (IDS). High accuracy and low False Alarm Rate could be pivotal challenges in developing IDS. To address this issue, this paper introduced an intrusion detection system by ensembling tree-based classifiers including decision tree, random forest and Gradient Boosted tree. The model is tested by different feature selection methods, and for evaluating its performance, the NSL-KDD dataset is applied. The results obtained show an improvement in accuracy in comparison with some existing methods.