Machine learning applications in mortgage default prediction

Abstract

Estimating default risk has been a major challenge in credit-risk analysis. Financial institutions are interested in the ability of a customer to payback a loan. In this research work we explore the application of some machine learning models in the prediction of mortgage defaults. We basically explore how machine learning methods can be used to classify mortgages into paying, default and prepay. This work examines the following machine learning methods: Logistic regression (simple and multi-class), Naive Bayes, Random forest and K-Nearest Neighbors. Finally, this work includes Survival analysis and Cox proportional hazard rate to estimate the probability of loan survival pass certain time and the impact of each variable in estimating the probability of survival respectively.