Bankruptcy Prediction of Indian Banks Using Advanced Analytics

The banking sector in India plays a crucial role in economic growth. A bank provides an opportunity for investments to encourage economic growth and the potential to yield higher returns. In this study, we develop a bankruptcy prediction model by using machine learning (ML) techniques, namely logistic regression, random forest, and AdaBoost, and compare these models with those developed using deep learning (DL) techniques, namely the artificial neural network (ANN). ANN results in the highest accuracy and the most favourable prediction model for bankruptcy. Data used in this study are collected from survived and failed private and public sector banks from India from March 2001 to March 2018. For bankruptcy prediction, we use the bank’s macroeconomic and market structure-related features. The feature selection technique ‘Relief algorithm’ is used to select useful features for the bankruptcy prediction model. Because failed banks in comparison with survived were less in the dataset, the issue of imbalanced cases may have arisen, in which case most ML and DL techniques do not perform well. Thus, we convert the dataset into a balanced form by using the synthetic minority oversampling technique (SMOTE). The results of this study can help in performing financial analyses of banks and thus have significant implications for their stakeholders.