An Empirical Analysis of Credit Risk Based on NonPerforming Assets of Selected Banking Sectors of India: Data Validation by Using Machine Learning Algorithms

Abstract

It was an attempt to study the impact of nonperforming assets (NPAs) in the selected public and private banking sectors from 2008 to 2019. An empirical study wasvalidatedby using machine learning (ML) algorithm models (Python via Jupiter Notebook, version 3.6) to know the credit risk. It was predicted the overall credit risk as per cut off GNPA>6, and GNPA>7through two types of models such as Regression and Classification models. In empirical findings, highly significant (p<0.001) change between studied banks as well as yearly data and GNPA & NNPA was recorded. Moreover,highly significant (p<0.001) differences were noted for the banking performance based on GNPA and NNPA and other macroeconomic variables viz. Unsecured/Tot Advances, GDP, CPI, Total Profit/ Total Advances, TL, GDP-1, Total Advances, RR, STA, Total Earnings/Total Advances, PSL, CPI-1.ForML study, the Naïve Bayes Classification was predicted to know how the Gross NPA is getting effected by different variables and obtained an accuracy of about 86% and the Support Vector Classification was obtained an accuracy of about 97% and about 100% for the Random Forest classifications, which seems like more realistic models.It may be varied with other independent variables like credit risk parameters and macroeconomic variables, etc. It is suggested in future to study with these cut off values for the determination of credit risk of these banking sectors. Keywords –Indian banking sectors; Machine learning algorithms; Non-performing assets; Empirical study