Using Machine Learning Techniques to Predict Financial Distress in Rural Banks in Indonesia

Abstract

LPS liquidated about 100 people's Rural Banks between 2015 and 2019, indicating that these banks are facing significant issues, particularly financial distress. This study seeks to forecast financial distress through a two-stage classification and regression approach. Researchers used financial report data from Rural Banks in Indonesia from 2015 to 2019, covering a total of 150 banks, with 50 financial ratios from bankrupt banks and 100 from those that remained operational. Data was analyzed for two consecutive years prior to any bankruptcy declarations. The classification targets are categorized into five classes: very healthy, healthy, quite healthy, unhealthy, and distressed. The study results demonstrate that the two-stage classification and regression method can effectively predict the onset of financial distress. This is validated by the classification outcomes using the Decision Tree Algorithm, which achieved an f1-score accuracy of 88%. The evaluation of timing predictions using Random Forest Regression revealed a mean absolute error of 1.2 months and a mean absolute percentage error of 3%. These predictions can assist regulators, bank management, and investors in making better-informed decisions to address financial distress risks in Rural Banks.  The superior performance of the Decision Tree Algorithm over Naïve Bayes in classifying financial distress highlights the potential of machine learning techniques in providing robust tools for early warning systems, aiding stakeholders in making informed decisions to mitigate risks.