Predicting corporate financial distress using KPCA and GA-based support vector machine

Abstract

In the analysis of predicting financial distress based on support vector machine (SVM), irrelevant or correlated features in the samples could spoil the performance of the SVM classifier, leading to decrease of prediction accuracy. On the other hand, the improper determining of two SVM parameters will cause either over-fitting or under-fitting of a SVM model. In order to solve the problems mentioned above, this paper used kernel principal component analysis (KPCA) as a preprocessor of SVM to extract the principal features of original data and employed the genetic algorithm (GA) to optimize the parameters of SVM. Additionally, the proposed GA-SVM model that can automatically determine the optimal parameters was tested on the prediction of financial distress of listed companies in China. Then, we compared the accuracies of the proposed GA-SVM model with those of other models of multivariate statistics (Fisher and Probit) and other artificial intelligence (BPN and fix-SVM). Especially, we adopted bootstrap technology to evaluate the reliability of validation. Experimental results showed that the GA-SVM model performed the best predictive accuracy and generalization, implying that the hybrid of GA with traditional SVM model can serve as a promising alternative for financial distress prediction.