Research On Economic Early Warning Methods Based On Support Vector Machine

Economic early warning is one of the most important research fields of economics. It is widely concerned about by all governments and public for its significance in economic subjects. The results of research on early warning show direct relation to the correctness of the cognition and judgment, to the choice of macroeconomic policies. However, usual warning methods often based on experts' experience or simple math models. And it is hard to deal with nonlinear problems so as not to meet the demand of macroeconomic early warning. As a popular arithmetic for data mining, Support vector machine or SVM has drawn much attention on this topic in recent years for its stabile basis in theory and good generalization.A macroeconomic early warning method is proposed by SVM combined with fuzzy theory and early warning research in this paper. Some new models are established to generalize and update SVM. Meanwhile, the methods are testified though data experiments in practice. Main results as following:1. Summarize the theories, research methods and developing history of economic early warning. Discuss the warning nature of usual early warning system including classical and new warning system and establish the frame pattern of warning system. Analyze the basic theory and character of Statistic Learning Theory (SLT) and SVM.2. Analyze the relations among pattern classification, SVM and macroeconomic early warning. Points out that early warning can be viewed as a process of pattern classification. For the first time a new intelligent warning system based on support vector classification is proposed which can auto-select the parameters in the model.3. It is required that every input must be exactly assigned to one of these two classes without any uncertainty in standard SVC. USVC early warning algorithm on expert advices is proposed for the first time, which is able to deal with the training data with uncertainty. Realize the effective combination between warning methods and expert intelligence.4. Ordinal support vector regression (OSVR) early warning algorithm is designed for multi-class early warning problem which label is associated with an integer from 1 to k. And fuzzy OSVR early warning Method is also designed, which can deal with the training data with uncertainty. A proper membership model named WBM is also proposed to fuzzify all the training data of every class.5. A new economic warning indices selection method is proposed for SVC early warning system based upon finding those indices which minimize bounds on the leaver-one-out error.6. Analyze the application of kernel function in SVM including support vector regression (SVR), kernel time series prediction and kernel principal component analysis (KPCA) comprehensive evaluation model.