Toward optimal bankruptcy prediction: evaluating ensemble methods using Taiwan and U.S. financial bankruptcy data

Abstract

Bankruptcy filings have increased significantly in many countries, causing widespread concern across society and triggering various economic issues. One contributing factor to the global rise in corporate bankruptcies is the unstable nature of companies’ growth. This issue often driven by unclear financial strategies and weak business direction. Thus, bankruptcy prediction plays a vital role, enabling earlier intervention and allowing business owners to improve their financial strategies proactively. This research investigates the effectiveness of ensemble machine learning (ML) methods using random forest (RF), stacking, and adaptive boosting (AdaBoost) for the prediction of corporate bankruptcy using datasets from Taiwan and the U.S. In the experimental phase, the performance is assessed using accuracy, precision, recall, F1 score, relative absolute error, and time. RF scored the highest accuracy in the classification of Taiwan’s bankruptcy data with 97.067%, meanwhile, AdaBoost M1 obtained the highest accuracy in the classification of the U.S.’s bankruptcy data with 94.0075%. The research shows that these methods, particularly AdaBoost M1, can improve early-warning systems and provide actionable insights for financial risk management. The main contribution of this research is its cross-country comparison of ensemble methods for bankruptcy prediction.