Credit Rating Analysis by the Decision-Tree Support Vector Machine with Ensemble Strategies

The recent financial tsunami and subprime crisis both triggered an excessive global financial decline. Since then, the ability to ensure the creditworthiness of firms, the development of a suitable credit rating mechanism and appropriate rating policies have become critical issues for most financial institutions. Credit rating is a multi-classification task. The decision-tree support vector machine (DTSVM) has been one of the most powerful models in dealing with multi-classification problems. However, the determination of important data features significantly affects the computation time and classification accuracy of DTSVM models. In addition, the inability to provide rules for decision-makers limits the practical applications of DTSVM models. This study proposes an M-DTSVM-RST model which integrates the unique strength of multiple feature selection strategies in feature determination, DTSVM in multi-classification, and rough set theory (RST) in rule generation. The advantages of the designed M-DTSVM-RST model include the ensemble learning ability in selecting essential attributes, the capability in yielding rules for decision-makers in multi-classification cases by the DTSVM technique with RST. Experimental results show that the M-DTSVM-RST model is a promising alternative in analyzing the credit rating problem.