Research of Bad Debt Risk Based on Rough sets and Binary Tree SVMMulti-layer Classifier

摘要

In this paper, a bad-debt-risk evaluationmodel is established based on rough sets and binary tree SVM multi-layer classifier. First, we use rough sets topre-process a new set of index system including typical 5Cfinancial indices system which combines both financial andnon-financial factors on the basis of the evaluation method. We define the bad debt rating as four classes- normality, attention, doubt and loss via analyzing accounts payable. Then, BP neural network is used to assess the 180 sampleswhich are stochastically extracted from listed companies, and the four classes are identified by the trained classifierusing 65 samples. Finally the binary SVM multi-layerclassifier is also used to compare the result with which from BP neural network. The test results show that the classifierhas an excellent performance on training accuracy andreliability. The experiment results indicate that multi-layer SVM classifier is effective in credit risk assessment andachieves better performance than BP neural network.