Support Vector Machines-Based Nonlinear Model And Its Application In Regional Economic Forecasting

With the fast development of machine learning and artificial intelligent algorithm in recent years, some new prediction technology is applied to economics and management research. Support vector machines (SVM) is a new method for pattern recognition based on the statistical learning theory, and it is one of the focuses of machine learning. SVM presents a lot of especial advantages for resolving the problems with small samples, nonlinear and high dimensional pattern recognition as well regression.Based on previous studies, this dissertation applied SVM to regional economic forecasting, due to SVM's high fitting precision, perfect generalization ability, global optimization and good compatibility with small samples. First of all, the evaluation index system of regional economic forecasting is proposed through quantitative and qualitative analysis, in regard to the change of policy. Secondly, given the economic development level in current period as the result of all kinds of factors in the past 3 years, all data samples are structured. Then this dissertation carries on the screening of the factors according to the correlativity analytical method. Moreover, during the optimization of model parameters, grid-searching algorithm is used to study the relationship between parameters and prediction accuracy of SVM, and the relationship among themselves. The parameter group with the best prediction accuracy for SVM is selected to set up the final model structure of SVM. Finally, through comparing the prediction accuracy of BP Neural Networks with that of SVM, it has proved the validity and superiority of SVM.This dissertation also discussed the effect of different kernel functions and data standardization methods on the prediction ability of SVM. By applying the presented method to economic forecasting of Chongqing, The results show that the SVM method provides excellent modeling and great generalization abilities when applied to a non-stable time series of regional economic with small data samples available. Accordingly, this method has some value of application, and it's considered to use SVM in the field of regional economic forecasting.