Predicting the Risk of MSME Bankruptcy in Kupang City: Integrating the Altman Z-Score, Macroeconomic Impact, Machine Learning, and Explainable AI (SHAP and LIME)

This study aims to assess the financial health and predict the bankruptcy risk of 100 micro, small, and medium enterprises (MSMEs) in Kupang City by integrating the Altman Z′′ score, testing the impact of macroeconomic variables, machine learning modeling , and explainable artificial intelligence (XAI). Data completeness was tested with Little's MCAR test, the impact of macroeconomics (inflation, deposit interest rates, and rupiah exchange rates) on the Z′′ score was tested with the Wilcoxon Signed-Rank Test, and four classification algorithms (Gradient Boosting, Random Forest, Logistic Regression, and Naive Bayes) were compared on 11 selected features, then the best model was explained with SHAP and LIME. The results showed that missing data was perfectly random (Sig. 0.587) and all MSMEs experienced a decrease in health scores after macroeconomic factors were included (Z = −8.697; Sig. 0.000). Gradient Boosting provides the best performance in accuracy, precision, recall , and F1-score (80%; 0.83; 0.71; 0.77) with an AUC of 0.85. SHAP analysis places D2, total current assets, and D1 indicators as global determinants, while LIME on three MSMEs shows that low D2 and limited total current assets drive the prediction of bankruptcy risk in MSME #2. These findings confirm that assessments based on internal data alone tend to be overly optimistic and that the combination of predictive models with transparent explanations can be the basis of an early warning system for businesses, banks, and local governments.