Purpose This study adopts a data-driven analytical framework to examine homestay pricing, combining machine learning models with model-agnostic explanation tools. Compared with traditional hedonic pricing approaches, it provides complementary and more transparent evidence on how facilities and services contribute to price formation, and offers preliminary insights into their implications for resource allocation. Design/methodology/approach Using data from 3,054 homestays in the Qiandao Lake area, this study constructs a three-tier feature system covering property attributes, facility configuration and service management. An XAI framework that integrates XGBoost with SHAP-based interpretability is then employed to quantify the dynamic contributions of these features to observed prices. Findings Significant nonlinear effects govern pricing. For example, a bimodal distribution emerges with respect to distance from attractions. SHAP-weighted facility/service indices outperform individual attributes in explaining price variation. High-value features drive price premiums, whereas excessive facility investment exhibits threshold effects and diminishing returns. Additionally, we conduct out-of-distribution robustness checks by extrapolating to a peak-season window with mean re-leveling and by re-estimating the core relationships via a generalized additive model. Both tests corroborate the transferability of the “Facility Index” and “Service Index” and the stability of key pricing mechanisms across seasons. Practical implications The facility and service indices translate complex model outputs into actionable levers. Hosts can prioritize high-impact amenity bundles, calibrate upgrade intensity to avoid over-investment and fine-tune seasonal price ladders for different property types. Destination managers can use the indices to benchmark homestay quality, anticipate price pressures in sensitive zones and design incentive schemes that align infrastructure provision, regulation and support with data-driven evidence. Originality/value This research pioneers the integration of explainable machine learning techniques with multidimensional indices, providing actionable strategies for resource allocation and dynamic pricing. The methodology offers novel theoretical insights for industry policy-making and future pricing research.
Zhang et al. (Tue,) studied this question.