Identification of Weak Banks Using Machine Learning Techniques: Evidence from the Indian Banking Sector

Early identification of ‘weak’ banks is essential for safety and soundness of the banking sector. Longer the delay in such identification, heavier the cost on an economy. In this article, we apply multiclass classification to classify banks operating in the Indian banking sector as ‘strong’ and ‘weak’ banks. Such classification is expected to provide direction to bankers for bank management. Using average return on equity (ROE) as basis for classification, we apply five machine learning models to the bank data set, namely Naïve Bayes, support vector machine, k-nearest neighbours, random forest and average neural networks. We find that all five models are able to predict the bank classes with a very high degree of accuracy. Ratio of non-performing assets to net advances turned out to be the most important variable in classifying banks as ‘weak’, followed by inflation and real exchange rate. The study is the first of its kind that successfully applies machine learning models in the Indian banking sector.