An empirical investigation of Lijiang’s residential property valuation in China using an integrated framework with Gaussian process regression and Bayesian hyperparameter tuning

Over the last decade, China’s property market has undergone pronounced cyclical shifts, transitioning from sustained growth to a marked slowdown following macroeconomic shocks in late 2021. These structural adjustments—shaped by evolving regulatory interventions and growing fiscal uncertainty—have posed complex forecasting challenges for stakeholders such as institutional investors, governmental regulators, and urban development planners charged with anticipating metropolitan housing value trends. To address the shortcomings of traditional prediction techniques, this research introduces an advanced framework based on Gaussian process regression, enhanced by adaptable kernel configurations and tailored feature engineering underpinned by domain expertise. The framework analyzes the temporal evolution of housing prices in Lijiang, Yunnan Province, utilizing a 9-year panel dataset spanning October 2015 to July 2024. By integrating Bayesian hyperparameter tuning with nested cross-validation, the model achieved substantial out-of-sample accuracy gains for October 2022 to July 2024, registering a relative root mean square error of 0.1209%. These results confirm the method’s ability to capture complex nonlinear dynamics and enrich spatial economic theory by revealing how local governance structures interact with price-setting mechanisms. Its modular architecture enables seamless coupling with additional econometric models or standalone applications across diverse urban contexts, offering scalable, evidence-based tools for policy decision-makers. Moreover, the findings provide strategic insights for designing adaptive regulatory frameworks, thereby linking this work to practical urban governance.