Interpretable-Machine-Learning-Driven Socio-Ecological Resilience Pathways in a Resource-Exhausted City: Evidence from Jiaozuo, China

Resource-exhausted cities face intertwined economic, social, and ecological pressures during transition, yet the dynamic evolution and pathway-specific responses of their socio-ecological resilience remain insufficiently understood. Using Jiaozuo, China, a nationally designated resource-exhausted coal-mining city, this study develops an interpretable-machine-learning framework that integrates resilience assessment, XGBoost-SHAP interpretation, spatial statistical validation, scenario simulation, and sensitivity analysis. A multidimensional resilience index was constructed for 2012–2022, and alternative development pathways were projected for 2030 and 2035. The results reveal stage-dependent resilience evolution, with model-explained drivers shifting from economy- and policy-related factors in 2012–2017 to a more ecology-oriented and multidimensional structure in 2017–2022. SHAP dependence and interaction analyses further identify nonlinear response patterns and conditional interactions among key social, economic, and ecological indicators. Scenario simulations show that green transformation produces the strongest model-predicted gains and remains the highest-ranked pathway under alternative subsystem-weighting schemes. These findings suggest that resilience enhancement in resource-exhausted cities depends on coordinated ecological restoration, industrial upgrading, economic vitality, and social adaptive capacity. The proposed framework provides a transferable approach for diagnosing resilience evolution and comparing transition pathways in resource-exhausted urban systems.