健康风险视角下我国潜在返贫风险人群影响因素探索及预测研究——基于CHFS数据

Exploration and Prediction Study of Influencing Factors of the Population at Risk of Potential Return to Poverty in China under the Perspective of Health Risk—Based on CHFS Data

摘要: 目的：识别潜在返贫风险高危因素，精准识别潜在返贫风险家庭可以为潜在返贫风险人群的早期识别、超前干预提供理论依据。方法：基于中国家庭金融调查数据库2019年数据，利用相关性分析以及LASSO回归结合多因素Logistic回归识别潜在返贫危险因素，利用十折交叉验证训练Logistic回归和XGBoost模型，比较两模型对于潜在返贫风险家庭的预测效能。结果：相关性分析与LASSO回归筛选变量后纳入17个潜在风险变量，多因素Logistic回归显示潜在返贫风险与健康状况、资产、收入、教育、户籍类型、大病保险、政府帮扶、脱贫户等因素有关。Logistic模型的F1值为0.609、AUC值为0.734，XGBoost模型的F1值为0.943、AUC值为0.948，XGBoost模型预测效能优于Logistic回归，收入、脱贫户、政府帮扶、户籍、存款是对模型贡献度最高的五个变量。结论：XGBoost模型能较为准确地预测潜在返贫风险，该模型能早期识别潜在返贫风险人群，为相关部门实施早期干预提供理论依据，助力全面推进乡村振兴。

Abstract: Objective: To identify high-risk factors associated with the potential risk of return to poverty; accurately identifying households at such risk provides a theoretical basis for the early detection of—and proactive intervention for—the affected population. Methods: Based on data from the 2019 China Household Finance Survey, this study employed correlation analysis, LASSO regression, and multivariate logistic regression to identify risk factors. Ten-fold cross-validation was used to train logistic regression and XGBoost models, and their predictive performance regarding households at risk of return to poverty was compared. Results: Seventeen potential risk variables were selected for inclusion following correlation analysis and LASSO regression. Multivariate logistic regression indicated that the risk of return to poverty was associated with factors such as health status, assets, income, education, household registration type, critical illness insurance, government assistance, and status as a household that had previously escaped poverty. The logistic model yielded an F1 score of 0.609 and an AUC of 0.734, whereas the XGBoost model achieved an F1 score of 0.943 and an AUC of 0.948; the XGBoost model demonstrated superior predictive performance. Income, status as a household that had escaped poverty, government assistance, household registration type, and savings were the five variables contributing most significantly to the model. Conclusion: The XGBoost model can accurately predict the risk of return to poverty. It enables the early identification of at-risk populations, provides a theoretical basis for relevant authorities to implement early interventions, and supports the comprehensive advancement of rural revitalization.