Consumer credit risk evaluation by logistic regression with self-organizing map

The significant growth of financial institutions´ concerns on consumer credit has resulted in a greater demand of classifying applicants into “good” and “bad” risk categories. Logistic regression has widely been employed as a mainstream approach to identify qualified consumers. The performance is, however, inevitably undermined by inappropriate/inconsistent samples. Self-organizing maps, which is known by its powerful unsupervised clustering capability, is investigated in this paper for its potential in credit risk assessment. An integration of self-organizing maps and logistic regression is proposed to firstly group consumers according to their featured attributes and then evaluate credit risk of consumers based on the logistic regressive model built on the group which the consumers most likely belong to. Besides, a confidence level weight is assigned to each model outcome. The performance of the integrated model is benchmarked by results of conventional logistic regressive model and local linear regressive model.