SVM Hyper-Parameters Optimization using Multi-PSO for Intrusion Detection

Abstract

Among all the classifiers available in recent time support vector machine (SVM) can be considered as one of the powerful classifiers. Support vector machine (SVM) modeled with properly tuned parameters can give significantly increasing accuracy rate for classification problem. Therefore, choosing the optimal values for the SVM hyper-parameters cannot be a mundane task. It is the most crucial and important task for SVM modeling. The basic emphasis of this paper is to optimize the values of C and γ which are two important kernel parameters of SVM which in turn can be used an intrusion detector in network. A set of properly optimized values of C and γ can increase the effectiveness and efficiency of SVM model. There are many state-of-the-art and metaheuristic techniques such as traditional grid search, gradient descent, genetic algorithm (GA), and particle swarm optimization (PSO). Among these, the most used optimization technique for SVM model selection is PSO. In our work, we have proposed a framework which uses a variant of PSO, multi-PSO for the selection of optimal values for the C and γ selection. The results show that it outperforms the other models for model selection in support vector machine.