Machine learning prediction of self-rated health among adults under different physical exercise participation statuses

Abstract

Self-rated health (SRH) reflects adults’ physical, psychological, and social well-being and is influenced by multiple determinants. Physical exercise is an important health-promoting behavior, yet evidence remains limited regarding the key predictors of SRH and their relative importance across different exercise participation groups. This study used machine learning to compare SRH prediction models across physical exercise participation, identify key predictors and feature-importance differences. Using data from the 2021 Chinese General Social Survey (CGSS2021), adults were divided into a Participate group and a Non-Participate group, and seven machine learning models, including Random Forests (RF), were developed to predict SRH status. Subsequently, the SHAP method was applied to quantify the contribution of individual features to model predictions. Among the Participate group, the RF model exhibited a relatively high AUC on the test set; amongst Non-Participate group, the LGBM model exhibited a relatively high AUC on the test set, with the performance differences between the primary models in the two groups being generally small. ‘Health Impact’, ‘Age’ and ‘Depression’ were common core predictors across both groups, although their ranking of importance and distribution of contributions were not entirely consistent. Variables such as ‘Internet Use’ and ‘Social Friends’ also demonstrated some auxiliary predictive value. Further stratification by physical exercise frequency and sensitivity analyses revealed that, although there have been changes in model performance and the ranking of some features, the core predictive factors have remained broadly stable. The key predictors of SRH among adults with different levels of physical activity participation share commonalities but also exhibit certain differences. Predicting SRH in adults requires a comprehensive consideration of multi-level information, including functional status, demographic characteristics, recent emotional experiences and social behaviors. This study contributes to a further understanding of the relative contributions and heterogeneity of predictors of SRH among individuals with different levels of exercise participation and may provide a reference for future health research and stratified health promotion. Given that this study utilized cross-sectional survey data from a single year, the external validity and practical applicability of these models require further validation.