Key predictors of subjective well-being among Chinese people: a multi-wave national cross-sectional analysis using machine learning, 2015–2023

Subjective well-being (SWB) refers to individuals’ subjective appraisal of their own well-being and provides invaluable insights for public policy. However, the relative importance of its numerous predictors remains poorly understood. This study employed a machine learning approach with multi-wave national data to identify the key predictors of SWB among Chinese people. Data were drawn from the 2015 ( N = 10,968), 2021 ( N = 5458), and 2023 ( N = 11,326) three-wave repeated cross-sectional surveys of the Chinese General Social Survey. Using the XGBoost algorithm, we modeled SWB scores based on 31 predictors. Analysis of feature importance consistently identified six factors as the most robust predictors across all three waves: perception of social equity, family economic status, depression, current social class, age, and self-reported physical health. The six-feature model achieved R 2 = 0.21–0.26 in development and 0.23 in external validation, recovering 84.0–86.7% of the variance explained by the full 31-feature model, demonstrating stable predictive performance across independent cohorts. These findings identify a parsimonious set of core predictors for the Chinese population and offer an evidence-based foundation for targeted policymaking. The simplified model also holds potential for practical application in policy simulation and decision support.