Explainable Sinkhole Susceptibility Mapping Using Machine-Learning-Based SHAP: Quantifying and Comparing the Effects of Contributing Factors in Konya, Türkiye

Sinkholes, naturally occurring formations in karst regions, represent a significant environmental hazard, threatening infrastructure, agricultural lands, and human safety. In recent years, machine learning (ML) techniques have been extensively employed for sinkhole susceptibility mapping (SSM). However, the lack of explainability inherent in these methods remains a critical issue for decision-makers. In this study, sinkhole susceptibility in the Konya Closed Basin was mapped using an interpretable machine learning model based on SHapley Additive exPlanations (SHAP). The Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM) algorithms were employed, and the interpretability of the model results was enhanced through SHAP analysis. Among the compared models, the RF model demonstrated the highest performance, achieving an accuracy of 95.5% and an AUC score of 98.8%, and was consequently selected for the development of the final susceptibility map. SHAP analyses revealed that factors such as proximity to fault lines, mean annual precipitation, and bicarbonate concentration difference are the most significant variables influencing sinkhole formation. Additionally, specific threshold values were quantified, and the critical effects of these contributing factors were analyzed in detail. This study underscores the importance of employing eXplainable Artificial Intelligence (XAI) techniques in natural hazard modeling, using SSM as an example, thereby providing decision-makers with a more reliable and comparable risk assessment.