The Prediction of Fall Risks in Older Adults: The Analysis of Machine Learning Algorithms and Feature Selections

Abstract Postural control becomes fragile by aging which increases the risk of falls, leads to fractures, and decreases the quality of life. It is important to investigate the fall risks in people with the older adults. In our study, we analyzed the KINECAL dataset which contains the records of 90 participants aged 18 to 92 years performing clinical movements. We performed six feature selection methods (the Information Gain, Fisher Score (GFS), Recursive Feature Elimination (RFE), Principal Component Analysis (PCA), Chi-square (CS), and Correlation-Based Feature Selection (CBFS)) and four classification algorithms (Random Forest (RF), Support Vector Machine (SVM), eXtreme Gradient Boosting (XGBoost), and Decision Tree (DT)). Our findings showed that XGBoost consistently outperformed the other algorithms, achieving the highest accuracy and F1 scores across most feature selection methods. XGBoost achieved its highest F1 scores with the original dataset (98.28%), CBFS (98.03%), and Information Gain (97.85%). On the other hand, Correlation-Based Feature Selection and Fisher Score generally achieved higher results among the feature selection methods. Conversely, the Support Vector Machine algorithm generally performed poorly. These findings strongly suggest that machine learning methods, particularly ensemble techniques like XGBoost, can be effectively utilized to predict fall risk, and they offer a foundation for developing new approaches in rehabilitation protocols. The study highlights the critical importance of selecting appropriate feature selection methods to improve the performance and interpretability of models used in rehabilitation.