基于机器学习的智慧医养融合平台用户行为分析与流失预警模型构建研究
Research on User Behavior Analysis and Churn Warning Model Construction of Smart Medical Care Integration Platform Based on Machine Learning

目的：探讨智慧医养融合平台的用户行为特征，构建基于机器学习的用户流失预警模型，为平台精细化运营提供决策支持。方法：以安徽省阜阳市某智慧医养融合平台2020年1月至2024年6月的用户数据为研究对象(n = 1237)，采用RFM模型分析用户行为特征，运用K-means算法进行用户分群，构建包括逻辑回归、随机森林、LSTM等多种机器学习模型进行流失预警。结果：用户以60~79岁人群为主(79.55%)，家政服务是最常用服务类型(50.62%)。识别出四类典型用户群体，其中低频–低价值用户的流失风险最高(风险指数为0.68)。LSTM模型展现出最优的预警性能(AUC = 0.961，准确率 = 0.923)，能够提前平均21天预警潜在流失用户。最近一次使用服务的时间间隔(SHAP值 = 0.284)是最重要的预测特征。基于模型预测结果的干预措施使高风险用户留存率提升32.5%。结论：基于机器学习的流失预警模型能有效识别高风险用户，结合用户行为分析的差异化干预策略可显著提升用户留存率，为智慧医养平台的运营优化提供了新思路。
Objective: This paper aims to explore user behavior characteristics of smart medical care integration platform and construct a machine learning-based user churn warning model to provide decision support for refined platform operation. Methods: Using user data (n = 1237) from a smart medical care integration platform in Fuyang City, Anhui Province from January 2020 to June 2024 as research subjects, the RFM model was applied to analyze user behavior characteristics. K-means algorithm was used for user clustering, and multiple machine learning models including logistic regression, random forest, and LSTM were constructed for churn warning. Results: Users were predominantly in the 60-79 age group (79.55%), with housekeeping services being the most frequently used service type (50.62%). Four typical user groups were identified, among which low-frequency-low-value users showed the highest churn risk (risk index was 0.68). The LSTM model demonstrated optimal warning performance (AUC = 0.961, accuracy = 0.923) and could predict potential churning users 21 days in advance on average. The time interval since last service use (SHAP value = 0.284) was the most important predictive feature. Intervention measures based on model predictions improved high-risk user retention rate by 32.5%. Conclusion: The machine learning-based churn warning model can effectively identify high-risk users. Differentiated intervention strategies combined with user behavior analysis can significantly improve user retention rates, providing new insights for the operational optimization of smart medical care platforms.