Unbalanced Prediction By SVM Of Enterprise’s Financial Distress Based On Over Sampling And Under Sampling

The development of human society based on the basic social organization which is the enterprise. The enterprise financial not only relationships in every aspect of the enterprise, but also is the most direct reflection of the results for business operations. So the early warning of financial crisis is very important e specially in the unfavorable economic development period. With the full passag e of the big data, Data mining has been applied in many fields. Accompanied by the capital market is becoming more and more standardized in China, artifi cial intelligence methods to predict corporate financial distress is one of thehot current research.Although there are more domestic and foreign scholars have been in finan cial distress prediction based on artificial intelligence methods. And it has form ed a complete system. But most of their research is based on the balanced fin ancial distress prediction. By observing the Chinese capital market and the rese arch in the past, this article’s Selection of financial distress based on non-equil ibrium view. Sme with the Previous studies. This article would Predict the Fin ancial Distress by improve the SVM in certain perspective of data,algorithm an d evaluation criteria. Compared with the single SVM classifier, the model Co nstruct in this paper is more effective.Firstly, from the perspective of data, we would introduced the SMOTE, To me Links and SMOTE+Tome Links. From the perspective of the algorithm, we would introduced the Bagging, Adaboost and SVM. From the perspective of ev aluation standards, we would use G and F. Model established in this paper is SMOTE+SVM, Tome Links+SVM, SMOTE+Tome Links+SVM, Bagging+SVM, Adaboost+SVM, SMOTE+Bagging+SVM, SMOTE+Adaboost+SVM, SMOTE+To me Links+Bagging+SVM and SMOTE+Tome Links+Adaboost+SVM.Secondly, in the empirical study, we collect 373 listed Companies’s financi al data which before the before the financial crisis two years between 2002 to 2012. And according to ratio of 1:3 to find 1119 normal companies. We use t he statistical methods to treat the data we have collected. Our experiments wer e based on the software named Matlab. Finally, we obtain result by compariso n, SMOTE+SVM is better than Tome Links+SVM and SMOTE; Adaboost+SV M is better than Bagging+SVM; SMOTE+Tome Links+Bagging+SVM is better than SMOTE+Bagging+SVM and SMOTE+Adaboost+SVM and SMOTE+Tome Links+Adaboost+SVM。we obtain that SMOTE+Tome Links+Bagging+SVM is t he best model when financial data is unbalanced.