Data Mining Techniques for Predicting the Non-performing Assets (NPA) of Banks in India

Banks in India are facing many challenges and witnessing many changes in recent times. Managing Non-Performing Assets (NPAs) has emerged as a major challenge for banks. This chapter presents the findings of a formal attempt to explain NPA variations from 2005 to 17. The findings are based on the application of various data mining techniques such as random forest, elastic net regression, and k-NN algorithm to understand the NPAs of banks in India. The study uses gross NPA as a dependent variable and other bank-specific and macroeconomic variables as independent variables. The experimental results show that elastic net regression is the best data mining technique to model the NPAs in the given context. Also, the empirical results in all the models have found strong evidence that certain variables like the previous year's NPA and the loan amount distributed have an impact on the NPAs. The findings of the study will provide policy directions to the banking sector and the government to control the quantum of NPAs in the financial system.