Dynamic Modeling on Credit Risk Evaluation with Fixed Time Window and Imbalanced Ensemble of Support Vector Machine
Qing-Hua Huang, Jie Sun and Wei-Dong Mao
Affiliation: School of Economics and management, Zhejiang Normal University, Ying Bin Da Dao 688, Jinhua City 321004, Zhejiang Province, P. R. China.
Commercial banks primary business is credit management. For commercial banks, how to make a scientific and effective credit decision is very important. Therefore, study on credit risk evaluation modeling has become an issue of great practical significance. For imbalanced sample data problem and concept drift problem which emerged in bank credit sample data with time passing, this paper proposed the use of rolling time window with a fixed window size and the ensemble of support vector machine (SVM) to simultaneously deal with concept drift and imbalanced dataset. Thus, we constructed a dynamic model on credit risk evaluation with fixed time window and SVM imbalanced ensemble. For concept drift, SVM was taken as the base model, rolling time window was applied to preliminarily process concept drift, and the ensemble mechanism was used to further improve the models capacity of dealing with concept drift in prediction. For imbalanced data, ensemble mechanism was proposed to improve the importance of minority samples and their influence in modeling process. The results of empirical studies on imbalanced samples of Chinese listed companies from 2001 to 2010 indicated that the dynamic model on credit risk evaluation with fixed time window and SVM imbalanced ensemble was able to continuously update new knowledge for prediction with time past and handle the problem of imbalanced data mining very well. This model had absolute advantages when comparing with static model with fixed time window and SVM ensemble. The review also discussed relevant patents about credit risk evaluation, the application of SVM, concept drift and fixed time window, which enriched the content of this paper.
Keywords: Concept drift, credit risk evaluation, fixed time window, imbalanced dataset, support vector machine ensemble, SAMPLE DATA, DYNAMIC MODELING, Base Prediction Model of SVM, EMPIRICAL RESULTS
Rights & PermissionsPrintExport