A Machine Learning-Based Fall Risk Prediction Model for Chinese Older Adults

OBJECTIVE: To predict the risk of falls in older Chinese adults, and to develop a machine learning model for assessing fall risk based on longitudinal study data, which can help identify high-risk individuals and provide a basis for formulating effective interventions. METHODS: A total of 2350 elderly people who participated in CHARLS during 2020 were selected for the study, and the prediction model was constructed. Several machine learning algorithms (logistic regression, K-nearest neighbor, random forest, neural network, decision tree) were used to assess fall risk. Explore the best cut-off points and adjust parameters in the training set, compare the prediction accuracy of the models in the test set, and further explain the best-performing models. RESULTS:Out of the 26 features analyzed, LASSO regression identified 12 features that were critical to model building. Comparing the performance of five different machine learning models on the test set, the results showed that the RF model performed best on various evaluation measures, with a threshold of 0.168, a specificity of 0.860, a sensitivity of 0.526, an accuracy of 0.693, an F1 score of 0.653 and an area under the ROC curve (AUC) of 0.738. .CONCLUSION:: The predictive model of machine learning can effectively evaluate the risk of falls in healthy elderly people. Random Forest (RF) methods clearly explain personalized fall risk prediction.