Support Vector Regression Based on Particle Swarm Optimization and Projection Pursuit Technology for Rainfall Forecasting

Accurate rainfall forecasting has been one of the most important role in order to reduce the risk to life and to alleviate economic losses by natural disasters. Recently, support vector regression (SVR) provides an alternative approach for developing rainfall forecasting model due to the use of a risk function consisting of the empirical error and a regularized term which is derived from the structural risk minimization. In this paper, a novel SVR model is presented for rainfall forecasting, which based on particle swarm optimization (PSO) and projection pursuit (PP) technology. First of all, we use the PP technology based on PSO to select input feature for SVR. Secondly, the PSO algorithm is used to search the parameters for SVR, and to construct the SVR models. Subsequently, the example of rainfall values in May in Guangxi is used to illustrate the proposed SVR--PSO--PP model. The empirical results reveal that the proposed model yields well forecasting performance, SVR--PSO--PP model provides a promising alternative for forecasting rainfall application.