This study aims to develop a model for predicting financial distress in Indonesian local governments, particularly to support the evaluation of forwarded foreign loans and strengthen early warning mechanisms. Unlike prior studies that focus on firms or national governments, this research proposes a cluster-based early warning model for local governments in an emerging economy.
Using a data set of 542 provincial and district/city governments from 2013 to 2022, this study applied machine learning techniques using Orange Software. Local governments were clustered based on Gross Regional Domestic Product expenditure and regional classification (Sumatra, Java, Kalimantan, Sulawesi, Bali and Nusa Tenggara, Maluku and Papua). Financial distress was defined using budget deficit thresholds and Debt Service Coverage Ratio, with financial and nonfinancial indicators serving as explanatory variables.
The empirical analysis identifies the most effective algorithms for each region and cluster using F1-scores, showing that machine learning models outperform traditional approaches in capturing fiscal vulnerabilities. Logistic Regression, Neural Network, Gradient Boosting, Decision Tree and Support Vector Machine yielded the best results, while Naive Bayes and Random Forest were not selected in any cluster.
The findings provide actionable guidance for designing early warning systems in sovereign risk management, enabling regulators and policymakers to strengthen fiscal monitoring, enhance loan approval accuracy and reduce default risk.
This research offers a novel regional and cluster-based approach to financial distress prediction in the public sector using machine learning. It contributes to accounting and finance literature by extending financial distress analysis beyond firms and national governments to local entities in a decentralized emerging economy.
