Regional Bank Personal Credit Rating Evaluation Based On Support Vector Machine

Under the impetus of economic development, market transaction becomes more and more diverse, such as mortgage and credit card, which are the important two forms. Because bank is the advance party of credit transactions, it is crucial to effectively grasp the personal credit situation and to avoid losses the interest for bank. In order to determine the quality of personal credit and their repayment ability accurately, the study of individual credit rating assessment becomes a priority issue in academic research. Traditionally, personal credit rating required setting scores of the various factors affected credit and summing up. The process is complex, time-consuming and can not guarantee its accuracy at the same time. With the development of computer technology, the classification algorithm has been applied to this issue, which has become a good solution to overcome the shortcomings of traditional methods exist.Currently, there are many methods used in personal credit evaluation system. However, most classification method has problems in the application process, including slow speed, inaccurate classification and so on. In addition, local differences lead to a unified approach for individual credit assessment. As a machine learning method, Support Vector Machine(SVM) becomes a good choice because this algorithm is based on statistical theory and it can improve the generalization ability, solve local optima according to the principle of structural risk minimization. Therefore, it is applicable in dealing with classification problems. Meanwhile, the personal credit rating is indeed the rating classification problems, so it will be a good idea if we utilize SVM for personal credit scoring system.Personal credit rating belongs to nonlinear classification problem and SVM method, in this paper, is applied to solve it for this feature. In order to obtain more perfect consequence, we improve the technology in allusion to the deficiencies of SVM method. First of all, we apply the principal component analysis to extract multi-dimensional information for data classifying and Coding. Through feature extraction technology, we can improve the training time and classification accuracy of SVM. Then, comparing with different experiments of kernel functions, we find it is obviously better if two or more kernel functions combined. Therefore, the establishment of SVM model combined principal component analysis with compounded kernel functions can effectively solve the problem of nonlinear classification.The experimental studies in this paper have shown that the reclassification of data we have already known, according to the optimization and solution of SVM, we can obtain similar results with the previous manual classification. At the same time, this approach has smaller manual intervention and is faster. In view of some progress in terms of personal credit rating evaluation, this study has value in terms of practical applications.