Network Anomaly Intrusion Detection System based on SVM and Gradient Boosted Trees

Abstract: Intrusion detection system (IDS) has recently become one of the fundamental parts of the security field. It is mainly comprised of two methods, namely anomaly detection and misuse detection. The focus of this paper is on a Network IDS (NIDS) based on feature selection by combining Support Vector Machine (SVM) and Gradient Boosted Trees algorithms. Different approaches have been used for increasing the accuracy to detect the intrusion. The first approach is the filter method using Fisher score and ReliefF score, the second one is the wrapper method and the third approach which brings novelty to this research is the combination of Fisher score and ReliefF score. However, the analysis of the technique is done using SVM with RBF-Kernel and Gradient Boosted Trees. This paper also includes Cross-Validation folds to perform a 10-flods Cross-Validation method for training and validation.

Keywords: Intrusion Detection System, Support Vector Machine, Gradient Boosted Trees, Feature ranking and selection.