Enhancing Human Fall Detection with Wearable Sensors: Exploring Artificial Intelligence through Approximate Entropy

Abstract

Falls among individuals living alone at home pose significant health risks, particularly for older adults, presenting a substantial challenge in healthcare. Detecting anomalies in their Activities of Daily Living (ADL) is crucial for proactive healthcare management, aiding in the early identification of potential issues and enhancing the quality of life for this demographic. Numerous studies have explored anomaly detection in ADL using various sensor types. Therefore, there is a pressing need to develop a precise system capable of accurately detecting human falls during ADLs within a home environment. This research specifically focuses on identifying and distinguishing human falls in ADLs by leveraging data collected from wearable sensors. This paper introduces a novel model employing the Approximate Entropy (ApEn) measure for detecting human falls during ADLs within home settings, achieving a notably high level of accuracy. The effectiveness of this proposed Approximate Entropy method is assessed using the publicly available URFD dataset. The experimental outcomes demonstrate that the entropy measures proposed exhibit promise in accurately detecting and distinguishing human falls from other activities. Comparative analyses with alternative techniques further corroborate the efficacy of the proposed Approximate Entropy.