Analysis of waist and wrist positioning wearable machine learning models to detect falls

Falls have a global impact, affecting people worldwide, with a notably high occurrence among the elderly. This study employs machine learning techniques to analyze falls and simulate Activities of Daily Living (ADL). The objective is to predict human falls by leveraging signals from accelerometers and gyroscopes as wearable sensors. By deriving statistical features such as mean, standard deviation, and range the authors successfully trained and assessed six machine learning models allowing them to compare solutions based on both wrist and waist data. The combination of these characteristics and sensors resulted in the Random Forest waist model achieving the most favorable metrics, with an accuracy rate of 97.22% in a 5‐s window. Falls are a widespread issue affecting people worldwide, regardless of their social status. This study presents various ML models, which can predict human falls using signals of a wearable sensor located on the wrist or the waist. The combination of these characteristics and sensors resulted in the RF waist model achieving the most favorable metrics, achieving an accuracy rate of 97.22%.