Optimal Wearable Sensor Placement for Human Fall Detection Using Machine Learning Models

Falls account for the major cause of injuries among the elderly and those employed in industrial settings, highlighting the necessity of a reliable and rapid fall detection system to ensure quick and timely medical responses, where wearable sensors play a key role, yet their ideal positioning on the body continues to be a research focus. In this paper, we used the public UP-fall detection dataset to investigate the optimal placement of wearable sensors for fall detection by leveraging machine learning models, including Random Forest (RF), XGBoost, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN). We evaluate sensor placements at five body positions—waist, pocket, ankle, wrist, and neck—using time windowed data (1s, 2s, 3s) at a sampling frequency of 18 Hz. Our results show that XGBoost consistently outperforms all other models used along with it, achieving a peak Fall F1-Score of 0.9907 with a 2-second window size when the sensor is placed in the pocket position. These findings will be of great value in optimizing where to put wearable sensors in the development of more effective and reliable fall detection systems.