Application of Machine Learning in Loan Default Forecasting

The growing importance of machine learning in decision-support systems has spurred substantial research into applying machine learning models to data from multiple sectors. Available open-source data on loan defaults has been used to train a variety of machine learning models from statistical models to complex neural networks. The performance of the models is evaluated using various information-theoretic and statistical measures. In addition to a detailed evaluation of the models, several elements associated with their practical adoption in a bank have been presented. Machine learning algorithms are now widely used and external advances in these methods have come, in large part, from academic research. That said, the focus for these models is now shifting to practical elements such as deployment, interpretability, regulatory compliance and operational aspects. To this end, data publicly available from the Lending Club, a leading US provider of loans to individuals and small businesses, has been used to develop different machine learning algorithms for prediction of the default of loans. The deployed solution uses tree-based methods both for interpretability as well as compliance wi