How can land development intensity and land use benefit be coordinated in high-speed railway station areas? Evidence from 1570 station areas in China

High-speed railway (HSR) station areas commonly face an imbalance of “high-intensity development and low-benefit output” during rapid construction, constraining their high-quality and sustainable development. To examine the coordination between land development inputs and land use outputs, this study investigates 1570 operational HSR station areas in China. Based on multi-source spatial data, a land development intensity–land use benefit (LDI–LUB) evaluation system is constructed, and the entropy-weighted TOPSIS model and coupling coordination degree (CCD) model are applied to measure coordination levels. A multi-scale “network–urban–place–node” indicator system is further developed. By combining XGBoost–SHAP interpretable machine learning with PLS-SEM structural path effect decomposition, this study reveals the multidimensional explanatory characteristics of CCD from predictive and structural association perspectives. The results show that China's HSR station areas are dominated by Low-efficiency development station area (LEDSA), with 970 station areas accounting for 61.8%, while High-efficiency development station area (HEDSA) includes 600 station areas, accounting for 38.2%. Compared with LEDSA, HEDSA shows a higher coupling coordination level, indicating a more stable synergy between land development inputs and land use outputs. XGBoost–SHAP results identify place-dimensional variables as the core predictive factors explaining CCD differences, especially the number of bus stops, road network length, and land use diversity, which show high predictive contributions, nonlinear responses, threshold characteristics, and interaction relationships. PLS-SEM further confirms that place-dimensional variables have stable positive structural associations in both types. Heterogeneity analysis reveals stage-specific associations among urban foundations, network locational conditions, and node attributes.