Corporate Financial Crisis Prediction Using SVM Models with Direct Search for Features Selection and Parameters Optimization

Since the accuracy of corporate financial crisis prediction is very important for financial institutions, investors and governments, many methods have been employed for developing effective prediction models. Support vector machines (SVM) are powerful methods for classification and have been used for this task. However, the performance of SVM is sensitive to parameters optimization and features selection. In this study, a new approach based on direct search and features ranking technology is proposed to combine features selection and parameters optimization for SVM models for financial crisis prediction. The sensitivity of features ranking technology, strategies of sampling training samples, and types of SVM models are analyzed on a data set with 2010 samples. The experimental results show that the proposed models are good alternatives for financial crisis prediction.