Parallel Multidimensional Step search algorithm for epsilon-insensitive support vector regression in time series prediction

Recently, Epsilon-Insensitive Support Vector Regression (epsiv SVR) has been introduced to solve regression and prediction problems. However, the preprocessing of data set and the selection of parameters can become a real computational burden to developer and user. Improper parameters usually lead to prediction performance degradation. In this paper, by introducing Parallel Multidimensional Step Search (PMSS) method, standard epsiv-SVR method is extended to a systematic approach for user to finish model selection with high prediction accuracy. Experiments with both simulation data set and practical data set were performed on computing nodes in Grid environment. Experimental results were analyzed with statistical method to validate the effectiveness and accuracy of the proposed method.