Abstract
In this study, we propose an automatic optimization least square support vector regression (LSSVR) using CPSO with a mixed kernel in order to solve regression problems. There are three parts for the LSSVR model. The first part is to consider the position of particles (solution) in chaotic sequence with good randomness and ergodic property of the characteristics in initial. The second part is the binary particle swarm optimization (PSO) employing to select possible input feature combination. Finally, a chaos search is used to select possible input features, and then we combine the optimize parameters optimized by PSO, called CP-LSSVR for short. For illustration and evaluation purposes, the CP- LSSVR is utilized to predict the remarkable datasets testing targets acquired from the UCI dataset. The results indicate that the proposed CP- LSSVR can produce a predicted model using a small number of features and show higher predictive capability than other methods listed in this paper.