Predicting Delinquency on Mortgage Loans: An Exhaustive Parametric Comparison of Machine Learning Techniques

Abstract

This paper explores the potential of 19 machine learning techniques to model and forecasts the risk of delinquency on mortgage loans. These techniques include variants of artificial neural networks (ANN), ensemble classifiers, support vector machine, K-nearest neighbors, and decision trees. ensemble classifiers variants. Our dataset comprises 14,062 mortgage loans that have been approved by bank underwriters in the US. We find that Multi-Layer Perceptron (MLP), a variant of ANN, outperforms all other techniques in training time and the precision for testing and training. We have also compared Artificial Neural Network-Multilayer Perceptron (ANN-MLP) results with the traditional binary logistic regression technique's findings. The comparison shows that the ANN-MLP behaves better than the binary logistic regression technique. The study suggests that ANN-MLP could be a valuable extension towards developing the existing toolkit, banks and regulators have to predict delinquency risk on mortgage loans.

Keywords

• delinquency risk prediction
• mortgage loan
• machine learning techniques
• artificial neural networks
• multi-layer perceptron
• binary logistics regression