Land subsidence is a significant geohazard in the Rhineland coalfields of Germany, primarily driven by large-scale open-pit mining and associated groundwater changes. This study integrates geospatial, geological, hydrological, and remote sensing datasets, including European Ground Motion Service (EGMS) measurements, to model subsidence susceptibility using a LightGBM classifier. The model demonstrates strong predictive performance in classifying four susceptibility levels (Low, Moderate, High and Very High). To enhance interpretability, Explainable Artificial Intelligence (XAI) techniques including PFI, LIME and SHAP were employed to identify key drivers of subsidence. Results consistently indicate that distance from mines and groundwater level are the dominant controlling factors, while faults and lithology provide secondary structural influence. Terrain and land cover variables contribute minimally. The integration of XAI proves critical in understanding model behavior, enabling not only accurate prediction but also transparent identification of underlying physical drivers. This improves confidence in the results and supports their application in risk assessment and mitigation planning. Field observations further validate that high-susceptibility zones correspond to areas of observed structural damage, confirming the reliability of the proposed framework.