A genetic programming-based credit risk assessment model

The acute necessity for evolving an effective and accurate credit default prediction model was felt post Global Financial Crisis of 2008. Financial institutions significantly revised and reformulated their risk management practices and gradually shifted towards machine learning-based credit risk management approach. Numerous machine learning-based models like logistic regression, artificial neural networks, decision trees, etc., are being employed by financial institutions for predicting the probability of default by the borrowers. In this paper, we introduce a genetic program (GP)-based model for predicting the probability of default and compare this model with other existing models in the domain of credit default and risk assessment. We used two different evaluation metrics for performance analysis: accuracy and negative log predictive density (NLPD) loss. Our results indicate that the proposed GP-based model has higher accuracy of prediction of credit default as compared to other risk assessment models.