Parameters Selection of Twin Support Vector Regression Based on Cloud Particle Swarm Optimization

Abstract

Twin Support Vector Regression (TSVR), a novel regressor, obtaining faster learning speed than classical support vector regression (SVR), has attracted the attention of many scholars. Similar to SVR, TSVR is also sensitive to its parameters. Therefore, how to select the suitable parameters has become an urgent problem for TSVR. In this paper, a parameters selection version for TSVR, termed parameters selections of twin support vector regression based on cloud particle swarm optimization (TSVR-CPSO), is proposed. Using the characteristics of randomness and stable tendency of normal cloud model, the inertia weight of PSO can be generated by the basic cloud generator of cloud model. To do so, we can improve the diversity of population for PSO, thus greatly improve the ability of diagnosis to avoid falling into local optimal. Based on the above idea, the cloud particle swarm optimization (CPSO) model is constructed. At last, CPSO is used to search the optimal combination of TSVR parameters. Simulations show that the proposed algorithm is an effective way to search the TSVR parameters and has good performance in nonlinear function estimation.