Machine Learning-Based Exploratory Analysis of Fall Risk Classification in Older Adults

This study investigates the application of machine learning models for classifying fall risk in older adults, with a focus on comparing the performance of two algorithms: Support Vector Machine (SVM) and K-Nearest Neighbors (KNN). The results demonstrate that KNN significantly outperformed SVM in classification accuracy, achieving 88.1 %, while SVM recorded a much lower accuracy of 47.6%. The statistical analysis revealed that KNN's higher accuracy is attributed to its ability to handle non-linear relationships between predictors more effectively, particularly when multiple clinical indicators are considered. Permutation feature importance analysis identified the Timed Up and Go (TUG) test and Hand Grip strength (HG) as the most influential factors, whereas the Functional Reach Test (FRT) contributed minimally. These findings suggest that KNN offers a more reliable model for predicting fall risk in clinical settings, enabling healthcare professionals to target interventions more accurately. Further research is recommended to validate these findings and explore additional predictive models for enhanced fall risk assessment in older adults.