Prediction modeling based on Bayes support vector machine

To solve the uncertainty, nonlinear and coupling problem of statistical data, Bayes support vector machine(BSVM) was proposed to predict their development trend. Herein, the uncertainty of data was described as BSVM weights with Gauss distribution. Based on the prior probability and Bayes theory, the parameters evaluation of BSVM was transformed into parameters optimization of posterior distibution, which can be obtained by the prior probability and Bayes theory. The nonzero vector as correlative support vector machine was selected, and the multiple dimension prediction model based on time serials and its parameters distribution were established. Considering the input of BSVM as random variable during every iterative process, the output of BSVM can be obtained with uncertainty transferring. Since BSVM can describe the influence of random variables and its tranferring, it can overcome the uncertainty and dependence influence and the prediction results approach to the real condition. Application indicates that the prediction of high-tech enterprise development based on BSVM can approach the actual condition with high precision and robust.