Abstract:

Aiming at the problems about the influence factor choosing, input character set optimizing and parameter optimizing of support vector machine (SVM) prediction, a whole process optimization method is proposed. Firstly, frequency pattern growth association rule is used to analysis and fuzzy Bayesian network combination model and solve the subjectivity problem in influence factor choosing. Then the fuzzy C means clustering algorithm is improved in dealing with the outlier and optimizing the distance in the clusters and among the clusters to get better input character sets, and this improved method reduces the redundancy and excesses of the training sample sets in the SVM forecasting model. Furthermore, the radial basis function is confirmed as kernel functions in the SVM model by comparing their characters. Modify the particle swarm optimization (PSO) algorithm about the particle speed, location and the inertia weight value and use the modified association PSO model to optimize the SVM parameters and build the forecasting model. Finally, the example shows the proposed method has more accurate prediction precision.