Prediction of Defaulters using Machine Learning on Azure ML

Abstract

Banks and lending institutions take risk in issuing new credit cards and loans to customers. Lending institutions at large need to have their own credit risk assessment system in accordance with Basel II guidelines. Many lending institutions lose a large amount of money as they do not have an accurate model to predict defaulters. The goal of credit risk management system is to accurately predict borrowers' ability to repay loans or make credit card payments in a timely manner. Researchers have taken multitude of approaches to solve this problem, and it continues to be an active area of research. Data mining and machine learning are emerging tools that are widely used by lending institutions to predict defaulters. These tools can effectively mine large dataset which is not feasible by traditional methods. In this work, we have used different algorithms including Deep Support Vector Machine (DSVM), Boosted Decision Tree (BDT), Averaged Perceptron (AP) and Bayes Point Machine (BPM) to build various models, in an attempt to better predict defaulters. Dataset, comprising of 25 attributes and 30k instances, was obtained from the repository of machine learning, University of California, Irvine (UCI). Our results show that, of all the four models used, DSVM can best predict defaulters. We believe that these models can be used to better predict defaulters by credit risk management system in banking and lending institutions.