AI Insight
This study used interpretable machine learning models to identify key predictors of subjective wellbeing among Chinese women using 2023 survey data. The XGBoost model revealed that perceptions of fairness had the strongest influence on wellbeing, followed by mental health, loneliness, physical health, and socioeconomic status indicators. The analysis uncovered nonlinear relationships between these factors, suggesting complex interactions that traditional linear models may miss.
Why it matters
The findings highlight that perceived fairness in society is a critical but often overlooked determinant of women's wellbeing in China, potentially more important than income or physical health alone. These insights can inform targeted policy interventions addressing social equity and mental health support for women, and demonstrate how machine learning can reveal nuanced patterns in social science research.
Understand the Science
BackgroundSubjective wellbeing (SWB) is an important indicator of women’s quality of life and social development. Identifying key predictors of women’s SWB is important for understanding their living conditions and informing policy-oriented research. Previous studies have often relied on traditional linear models, which may not fully capture the multidimensional and nonlinear nature of women’s SWB.MethodsUsing data drawn from the 2023 Chinese General Social Survey (CGSS), this study applied an interpretable machine learning framework to predict SWB among Chinese women from fairness, health, and socioeconomic status (SES) indicators. Five models—Elastic Net, Random Forest, XGBoost, LightGBM, and an artificial neural network (ANN)—were trained and evaluated. SHapley Additive exPlanations (SHAP) was used to interpret the final model.ResultsXGBoost and LightGBM showed highly similar predictive performance, and XGBoost was selected for SHAP interpretation based on cross-validation (CV) performance. SHAP analyses showed that fairness made the largest contribution to model-predicted SWB, followed by mental health, loneliness, physical health, and subjective SES indicators. SHAP dependence plots suggested nonlinear and possible interaction-like predictive patterns involving fairness, future subjective SES, personal income, and physical health.DiscussionThis study demonstrates the usefulness of interpretable machine learning for examining women’s SWB and provides model-based evidence on its multidimensional predictive structure among Chinese women. The findings may inform future policy-oriented, longitudinal, and causal research on social fairness, health, SES, and women’s wellbeing.