Forecasting residential property values in Jinhua City, China: a machine learning framework integrating Gaussian process regressions

Purpose The Chinese real estate sector underwent a period of rapid growth during the previous decade, followed by a contraction phase commencing in late 2021. This structural transition, precipitated by dynamic macroeconomic shifts, has introduced significant complexities in property valuation forecasting for both regulatory authorities and market participants. The purpose of this study is to demonstrate the building of a machine learning model for addressing the complicated property price forecasting problem. Design/methodology/approach In response to these methodological challenges, this investigation uses Gaussian process regression (GPR) incorporating multiple kernel configurations and basis functions to systematically examine monthly residential price fluctuations in Jinhua City, Zhejiang Province, using a comprehensive data set spanning April 2015 through July 2024. Findings The modeling framework was developed through Bayesian hyperparameter optimization and rigorous cross-validation techniques, demonstrating robust predictive accuracy during the out-of-sample evaluation period (September 2022–July 2024), as evidenced by a relative root mean square error metric of 0.1165%. Originality/value The empirical outcomes provide substantive foundations for enhancing theoretical models of urban housing market dynamics while offering practical insights for policy formulation, either as standalone analytical tools or when integrated with supplementary predictive frameworks.