Utilizing SMOTE-TomekLink and machine learning to construct a predictive model for elderly medical and daily care services demand

This study aims to construct a prediction model for the demand for medical and daily care services of the elderly and to explore the factors that affect the demand for medical and daily care services of the elderly. In this study, a questionnaire survey on the demand for medical and daily care services of 1291 elderly was conducted using multi-stage stratified whole cluster random sampling. SPSS21.0 statistical analysis software was used to describe the basic data of the elderly statistically, and univariate analysis was used to screen variables for model construction and binary logistic regression analysis. The acquired dataset has class imbalance, and to handle this issue, Synthetic Minority Over Sampling Technique with TomekLink (SMOTE-TomekLink) was adopted to resample the dataset for class-balancing. To improve computational efficiency, we used three algorithms to develop prediction models, including Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Light Gradient Boosting Machine (LightGBM) algorithms. The performance of each model was measured, and the performance of the prediction model was obtained using the following performance metrics: accuracy (ACC), recall (R), precision (P), F1-score, and area under the receiver operating characteristic (AUC). The prediction models for the medical and daily care services demand of the elderly were developed and validated using 12 and 13 key features, respectively. The LightGBM algorithm emerged as the superior prediction model for estimating the service needs of the elderly. For the medical service demand prediction model, LightGBM achieved an AUC of 0.910 and F1-score of 0.841. In the daily care services demand prediction model, LightGBM demonstrated an AUC of 0.906 and an F1-score of 0.819. In the LightGBM model, the analysis of feature importance indicates that the number of chronic diseases, education level, and financial sources emerge as the most significant predictors for the demand of healthcare services, encompassing both medical and daily care services. Based on questionnaire information combined with feature selection, unbalanced data processing and machine learning methods, this study constructed a machine learning model for predicting the demand for medical and daily care services for the elderly, and analyzed the influencing factors of the demand for medical and daily care services for the elderly, providing a reference for the construction and verification of future prediction models for the demand for medical and daily care services for the elderly.