Predicting educational loan defaults: application of artificial intelligence models

Abstract

We show that Educational loans is a case for application of artificial intelligence models to predict potential defaulters with a reasonable accuracy. Ensemble models tend to perform better than simple artificial techniques and statistical models and that the performance can be improved significantly by model stacking. We argue here that a stacked model created using a few sparsely correlated base models is likely to be the best model for predicting Educational loan defaults given that the interaction between diverse features would create non-linearities that are impossible to model using a single model, there is little a priori knowledge of the distribution of educational loan defaults and the relationships between various factors that govern the distribution. It is evident that collateral-free loans have a considerably higher rate of default with moral hazard problem as compared to the loans with collateral. Students qualifying from well rated educational institutions are prone to strategic default or wilful default. Considering the impact of macroeconomic conditions greatly improve the classification accuracies