Explainable Artificial Intelligence (XAI) to interpret the contributing factors of land subsidence susceptibility prediction model

• Explainable AI (XAI) used to explain the complex phenomena of land susceptibility by determining the best contributing factor that influence the most. • Three ensemble model; Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Extra Trees (ETR) were used to predict land subsidence based on eleven (11) contributing factors. • Random Forest (RF) is the best ensemble model to predict potential areas that prone to land subsidence. • Shapley Additive exPlanations (SHAP) explain the predictions of the ETR, RF, and XGBoost models and justify the contribution of each factor toward land subsidence susceptibility. • Distance to road and elevation highly influence land subsidence susceptibility. Land susceptibility prediction is crucial for mitigating natural disasters and optimizing land use planning. While powerful, traditional black-box machine learning models often lack interpretability, making it difficult for stakeholders to trust and understand their predictions. This study explores applying Explainable Artificial Intelligence (XAI) techniques to demystify the contributing factors in land susceptibility prediction models using Extra Tree, Random Forest, and XGBoost regression. We employ grid search tuning to compare these models and identify the one that fits our data. Additionally, we leverage Shapley Additive exPlanations (SHAP) to provide comprehensive insights into model behavior and decision-making processes. By applying XAI to the historical land susceptibility dataset, we identify key factors such as elevation, slope, aspect, slum areas prone areas, groundwater, topography wetness index (TWI), distance to road and rivers, land use, and soil types that have significance influence on land susceptibility predictions. Results show that topography elevation and distance to road influence land susceptibility more than other factors. This study demonstrates that integrating XAI methodology can enhance model performance and help in the effective communication of findings to policymakers and land managers. This approach not only improves the interpretability of susceptibility predictions but also fosters better-informed decision-making for anomaly detection, land use management, and disaster risk reduction, especially for areas near critical infrastructures.