A Hybrid Model Based on ANFIS and Nonlinear Feature Selection for Credit Risk Evaluation

Abstract

Credit risk evaluation is an important decision process to financial institutions. Feature (variable) selection is a key step to many credit evaluation problems and it is often used as a dimension reduction technique to process credit data. However, the traditional correlation-based feature selection (CFS) is a linear analysis method when calculating the correlation coefficient and it cannot deal efficiently with nonlinearly correlated variables. This paper presents an improved approach of nonlinear correlation-based feature selection—Gebelein’s maximal correlation-based feature selection (GCFS), based on analysis of CFS and Gebelein’s maximal correlation (GMC), to realize the data reduction. Furthermore, an integrated model, GCFS-ANFIS model, is presented combined GCFS with Adaptive Neuro Fuzzy Inference System (ANFIS). The proposed model has been applied to credit evaluation based on the data collected from a set of Chinese listed corporations, and the results indicate that the performance of the GCFS-ANFIS model is much better than the ones of the other classic methods.