Identifying and ranking the predictors of social insurance participation among informal workers: An interpretable machine-learning analysis of China's pension and medical schemes

Voluntary social insurance is a key instrument for extending protection to informal workers, yet participation remains low. This study identifies and ranks the predictors of informal workers' participation in China's voluntary pension and medical insurance schemes. Using panel data from the 2018, 2020, and 2022 waves of the China Family Panel Studies (CFPS), we apply machine-learning models and Shapley additive explanations (SHAP) to analyze high-dimensional data and generate an interpretable ranking of associated factors. Risk-related factors dominate the prediction of pension insurance participation, accounting for 43.1% of total SHAP importance, followed by institutional factors at 18.3%. Medical insurance participation exhibits a more balanced predictive structure: individual characteristics account for 22.4%, followed by risk factors at 19.4% and institutional factors at 18.9%. Intertemporal analysis further shows that income stability is strongly associated with continued participation in both schemes, whereas rising medical expenditure is one of the strongest predictors of continuous medical insurance coverage. By systematically classifying and ranking multidimensional predictors through interpretable machine learning, this study provides empirical evidence relevant to the redesign of voluntary social insurance aimed at mitigating adverse selection and improving coverage among informal workers.