An Interpretable Personal Credit Evaluation Model

Abstract

How to establish a personal credit evaluation model with both interpretability and high prediction accuracy is an essential task in the credit risk management of commercial banks. To realize interpretable personal credit evaluation with high accuracy, it proposes an interpretable personal credit evaluation model DTONN (i.e., Decision Tree extracted from Neural Network) that combines the interpretability of decision tree and the high prediction accuracy of neural network. Significant features were selected from raw features by a decision tree, and a four-layer neural network was constructed to predict personal credit by using the selected features. Therefore, the accurate credit evaluation was made through the neural network and associated decision process was intelligibly displayed in the form of a decision tree. In the experiments, DTONN was compared with four personal credit evaluation models: decision tree, neural network, support vector machine, and logistic regression, on give-me-some-credit credit dataset. The experimental results show that our proposed model is state-of-the-art both on the accuracy and interpretability.