High-Risk Population Identification Based on Machine Learning

Abstract

This research centers on identifying high-risk populations with poor digital health capabilities among older adults in Beijing in the post-pandemic context. The goal is to identify high-risk populations precisely by comparing various machine learning algorithms, offering support for public health policymaking.The study adopts the social ecosystem theory to develop an indicator framework, collects data through a mix of online and offline approaches, and performs empirical analysis using SVM, Random Forest, and XGBOOST classification algorithms with L-SMOTE interpolation resampling to compare their effectiveness in identifying high-risk populations. Results indicate that the SMOTE-resampled SVM algorithm outperforms in accuracy, recall, and F1 score, while the L-SMOTE-resampled XGBOOST algorithm also demonstrates solid performance. Thus, when considering recall, precision, and F1 score comprehensively, the SMOTE-resampled SVM algorithm provides the best prediction results for high-risk populations.