Explainable machine learning and GIS analytics for sustainable tourism planning: a destination decision-support approach

Purpose This study aims to develop an analytically grounded destination decision-support framework that integrates GIS-based spatial analysis with explainable machine learning to identify tourism supply-demand gaps and assess suitability for national-scale 5A scenic-spot cultivation screening. Design/methodology/approach A multi-threshold two-step floating catchment area (2SFCA) method was used to identify tourism gaps at the 50 × 50 km grid scale. Eleven variables representing tourism resource endowment, public service facilities, transport accessibility and ecological quality were used in four machine learning models. Cross-validation selected the primary suitability model, and ecologically constrained grids were excluded using institutional nature reserve boundaries. Shapley additive explanations (SHAP) and GeoSHAP were used to identify key drivers and reveal nonlinear effects and spatial heterogeneity. Findings Tourism gaps account for about 20% of grid cells and show stable clustering across distance thresholds, mainly in the Tibetan Plateau, the northern continuous belt and the southwest plateau mountains. XGBoost was retained for suitability prediction and interpretation. Tourism attractions and trunk accessibility were the dominant drivers, jointly contributing over 61%. Tourism attractions show pronounced spatial non-stationarity, whereas trunk accessibility has weaker spatial moderation. After applying ecological constraints, only a limited number of high-suitability grids with existing 4A scenic spots remained. Originality/value By integrating tourism gaps, institutional ecological constraints and explainable machine learning within an analytically grounded decision-support framework, this study improves the transparency of national-scale priority-area screening for sustainable tourism planning.