Fall Risk Prediction for Elderly Using Head-Mounted Inertial Sensors and Tree-Based Models
Yu-Zheng Chen, Fang-Yi Lin, L Tseng, Chien‐Hsu Chen, Pi‐Shan Sung, Chih‐Lung Lin · 2024
In recent years, with the increasing number of elderly people, falls have caused significant concerns and troubles in their daily lives. Therefore, the importance of fall prediction is indispensable. This work applies a head-mounted device with an inertial measurement unit, and tree-based models are used to predict the future fall risk of elderly individuals. Elderly participants were asked to stand in place with their eyes closed for 1 minute. A random forest and Morse fall scale were utilized to select features of acceleration and angular velocity from time-domain and frequency-domain. Because the decision tree is more lightweight than the random forest, the decision tree can be more suitable for use on edge devices. Therefore, a decision tree was utilized to classify the future fall risk of elderly individuals. With the optimal feature subsets, the 10-fold cross-validation results yield recall, precision, and F1-score of 0.818, 0.692, and 0.750, respectively.