A novel weighted support vector machines multiclass classifier based on differential evolution for intrusion detection systems

This study compares several methods for creating a multiclass, support vector machines-based (SVM) classifier from a set of binary SVM classifiers. This research aims to identify multiclass SVM models best suited to the intrusion detection task. The methods we compare include one-against-rest SVM (OAR-SVM), one-against-one SVM (OAO-SVM), directed acyclic graph SVM (DAG-SVM), adaptive directed acyclic graph SVM (ADAG-SVM), and error-correcting output code SVM (ECOC-SVM). We also propose a novel approach, based on weighted one-against-rest SVM (WOAR-SVM). Using a set of meta-heuristically generated weights, a WOAR-SVM model is able to compensate for errors in the predictions of individual binary classifiers. In addition, this approach enables seamless integration of several binary hypotheses into a composite, multiclass hypothesis, where each binary classifier may feature a unique set of classification parameters. The results of our experiments on the NSL-KDD benchmark dataset for IDS indicate that WOAR-SVM outperforms the other approaches in terms of overall accuracy.