Credit Risk Assessment: An Active Learning Approach

J. Wu and X. Zhang (PRC)


Credit risk assessment, active learning, machine learning


Modern credit risk management aims at assessing the default probability (DP) of a debtor according to his historical and current financial data. Due to its prominent importance in credit loan decisions, the DP assessment problem becomes a research focus in the field of financial mining. Previous methods for the problem mainly build a DP assessment model passively from the available training data and do not make use of abundant unlabeled data. In this paper, however, we propose an active learning approach, in which a model is built with the interaction of humans. Specifically, our model consists of two sub-models which predict a debtor’s DP from different mathematical aspects. Additionally, our model actively selects an unlabeled instance for humans to label (i.e. give its DP) according to its information density as well as the disagreement of the two sub-models on its label. We also develop a special method to measure the information density of an unlabeled instance. Experimental results reveal that the active learning approach performs much better than its passive learning version.

Important Links:

Go Back