a synergistic hybrid ensemble model for predicting financial distress

MBEMBA BADJIE, Fitri Amalia, S.E., M.Sc., Ph.D.

2026 | Tesis | S2 Ilmu Akuntansi

Financial distress prediction remains a critical challenge in credit risk management, while traditional prediction models often exhibit limited predictive accuracy. This study integrates ensemble learning techniques with macroeconomic variables to enhance financial distress prediction.

The study aims to develop and validate a stacking hybrid-ensemble model for predicting financial distress using 993 firm-year observations from three countries: the United States, China, and Indonesia.

The proposed model employs Random Forest, Gradient Boosting, and Multilayer Perceptron (MLP) as base learners, with Logistic Regression serving as the meta-learner. Temporal validation was implemented by training the model on data from 2015–2019 and testing it on data from 2020–2024. Model performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC-ROC), sensitivity, and specificity.

The stacking hybrid-ensemble model achieved an AUC-ROC of 0.9145, with a sensitivity of 88.68% and a specificity of 91.52%. The inclusion of macroeconomic variables improved predictive performance by 1.60% compared with models using only financial variables. Statistical significance was confirmed (p < 0>

The proposed ensemble model provides accurate and robust financial distress predictions across diverse economic environments. Furthermore, the integration of macroeconomic variables and country-specific decision thresholds enhances the model's practical applicability for global credit risk management.

Financial distress prediction remains a critical challenge in credit risk management, while traditional prediction models often exhibit limited predictive accuracy. This study integrates ensemble learning techniques with macroeconomic variables to improve financial distress prediction.

The study aims to develop and validate a stacking hybrid-ensemble model for predicting financial distress using 993 firm-year observations from three countries: the United States, China, and Indonesia.

The proposed model employs Random Forest, Gradient Boosting, and Multilayer Perceptron (MLP) as base learners, with Logistic Regression serving as the meta-learner. Temporal validation was implemented by training the model on data from 2015–2019 and testing it on data from 2020–2024. Model performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC-ROC), sensitivity, and specificity.

The stacking hybrid-ensemble model achieved an AUC-ROC of 0.9145, with a sensitivity of 88.68% and a specificity of 91.52%. The inclusion of macroeconomic variables improved predictive performance by 1.60% compared with models using only financial variables. Statistical significance was confirmed (p < 0>

The proposed ensemble model provides accurate and robust financial distress predictions across diverse economic environments. Furthermore, the integration of macroeconomic variables and country-specific decision thresholds enhances the model's practical applicability for global credit risk management.

Kata Kunci : Financial distress prediction, stacking ensemble, machine learning, macroeconomic variables, credit risk management, temporal validation