Machine Learning Approaches for Lightweight, Reliable, Generalizable, and Explainable Classification of Accelerometer-Based Measurements of Movements and Falls

This chapter presents and discusses the desired characteristics that the research community and users expect from machine learning (ML) approaches for the classification of motion-based activities of daily living (ADLs) using accelerometer measurements, with an emphasis on recognizing various types of falls. Falls are a major contributor to injuries and fatalities among the elderly population, highlighting the importance of real-time fall detection and alert systems. ML algorithms have demonstrated high accuracy in fall detection in controlled experimental environments. However, their effectiveness in real-world situations requires further investigation.