Multi-Objective Hyper-Heuristic Improved Particle Swarm Optimization Based Configuration of Support Vector Machines for Big Data Cyber Security

Abstract

The massive increase of information in the big data era has not only created data processing problems, but also the data security issues. These big data cyber security issues can be handled effectively using machine learning algorithms among which the Support Vector Machines (SVM) has better results on big data classification problems. Defining the proper configuration of the SVM requires expert knowledge in selecting the kernel function and other parameters and this can significantly improve its classification results. In this paper, the SVM configuration process is modelled as a multi-objective optimization problem by considering the false positive rate, false negative rate and model complexity parameters. A Hyper-Heuristic Improved Particle Swarm Optimization (HHIPSO) framework is developed to optimize the SVM multi-objective optimization problem by incorporating the hyper-heuristics and improved particle swarm optimization algorithm. The proposed hyper-heuristic framework includes the high-level strategy for controlling the selection of low-level heuristics by search process and the low-level heuristics generate the new SVM configuration solutions using different rules of PSO. The effective selection of the kernel function and the respective parameters of the SVM should result in better values of false positive rate and false negative rate and also reduce the complexity. The evaluation of the proposed HHIPSO is performed on two cyber security problems and the obtained results illustrated that the proposed approach is effective in improving the classification of big data cyber security problems than the other algorithms.