A Comparative Evaluation of Machine Learning Models for Detecting Human Falls Using Wearable Sensors

The paper provides a comprehensive comparative analysis of geriatric fall detection utilizing multiple machine learning algorithms, addressing the essential topic of fall prevention in older persons. The SisFall dataset contains wearable sensor data, which is preprocessed to ensure compatibility and robustness of the model. To capture all essential data, motion characteristics are retrieved in both the frequency and temporal domains. Multiple machine learning approaches are used for classification, which include(s) Support Vector Machine (SVM), k-Nearest Neighbors (k-NN), Extreme Gradient Boosting (XGBoost), Random Forest (RF) and 1D Convolutional Neural Networks (1D-CNN). Model performance is measured using k-fold cross-validation, which includes measures such as specificity, sensitivity, and accuracy. Ensemble approaches such as RF and XGBoost achieve the maximum accuracy at 98.21%. The findings highlight the potential of machine learning models and wearable sensors to enhance elder safety by providing reliable fall detection and reducing the risk of falls.