Advancing Fall Detection with Machine Learning: A Review of Fall Risk Assessment Methods and Implementation on Wearable Devices

Falls among the elderly population have become a significant healthcare concern, leading to middle to severe injuries and placing a substantial burden on healthcare systems. Mitigating this problem requires effective fall risk assessment and timely fall detection. This thesis aims to investigate methods for continuous fall risk assessment and fall detection using a wearable device. The first objective is to review existing literature on fall risk assessment and fall detection methods. This review will inform the selection of suitable algorithms, considering technical feasibility and accuracy. Specifically, the study will explore whether machine learning (ML)-based models outperform classical methods when utilizing an inertial sensor. The second objective is to develop a fall detection algorithm, employing various machine learning models and classical methods to accurately detect falls in real-time. The algorithm will be tested on a representative dataset, with an analysis of potential limitations. This research aims to contribute to the advancement of fall detection algorithms, ultimately reducing the impact of falls on the elderly population.