Deep Learning-Based Early Warning System for Bankruptcy Risk in Indian MSMEs: A Feasibility Study

The objective of our work is to present a feasibility study on developing a deep learning-based early warning system for bankruptcy risk in Indian SMEs. The study explores the use of various deep learning models such as Back Propagation Neural Networks (BPNN) and Radial Basis Function Neural Networks (RBFNN) in addition to traditional machine learning models such as Gradient Boost classifier and Decision Trees to predict the probability of bankruptcy of Indian SMEs. The data was procured from the Centre for Monitoring Indian Economy (CMIE) Database for MSMEs from 2015 to 2021. We used Backward Feature Elimination and SMOTE for feature selection and data pre-processing, respectively. We used ratios like debt-to-equity ratio to reduce heteroskedasticity in our models. We found machine learning models performing better than deep learning models for predicting Bankruptcy in Indian SMEs. They perform better with tabular data and are significantly faster than deep learning models. The early warning system should help stakeholders, including investors, lenders, and insurers, to eliminate bad financial conditions for SMEs and take crucial steps to prevent financial loss. These studies contribute to business threat management and address the power of machine learning models in predicting the threat of bankruptcy in the SME sector in India.