<![CDATA[This paper examines the effectiveness of an explainable machine learning–based Early Warning System (EWS) in detecting financial distress in Islamic banking. Following recurrent episodes of financial instability, regulators and bank managers increasingly require predictive tools that are not only accurate but also interpretable. This study aims to develop a transparent and reliable distress prediction framework for Indonesian Islamic banks. Conventional early warning models can signal which banks are at risk but fail to explain why distress occurs at the variable level, limiting their usefulness for supervisory intervention. This study introduces an integrated Explainable Artificial Intelligence (XAI) framework that combines ensemble tree-based machine learning with SHAP and LIME to uncover non-linear threshold effects in the risk–stability relationship an aspect largely overlooked in prior Islamic banking studies. Using panel data from Indonesian Islamic banks over the period 2014–2024, financial distress is modeled as a binary outcome based on asset quality and profitability thresholds. Predictive performance is evaluated using Random Forest algorithms and benchmarked against conventional logit models, while interpretability is achieved through SHAP and LIME analyses. The results show perfect predictive separation (AUC = 1.00), with operational inefficiency (BOPO), profitability (ROA), and asset quality (NPF Gross) emerging as dominant risk drivers, exhibiting clear non-linear threshold effects. The study concludes that explainable machine learning significantly enhances both predictive accuracy and supervisory interpretability, offering a robust and policy-relevant tool for early intervention in Islamic banking stability.]]>
Copyrights © 2026
