Improving retail banking loans recovery via data mining techniques: a case study from Indian market

Abstract

In 2006-2007, the Indian banks saw a phenomenal increase in their loans, because of global growth, and mortgage market in the USA. But this was a 'bubble', hence did not sustain. Then global recession set in affecting the financial market in India. The default rates on unsecured borrowing rose and recovery became difficult. Banks spent more resources for their recovery. But in the process, borrower information was ignored, although credit bureau information about the borrower was available. This paper demonstrates that data mining techniques can find out defaulters who are most likely to pay, hence focusing recovery efforts on them. We tested the predictive power of neural network (NN), CART (DT) and logistic regression (LR) on the data of one of the bank's personal loan portfolio. Also, we demonstrated the use of 'textual data' available in the form of interaction with the borrowers and its value addition in predicting their payment behaviour.