A Machine Learning Approach for Near-Fall Detection Based on Inertial and Force Data While Using a Conventional Rollator

Abstract

Falls are a major concern for society. They may result in death or in several injuries that require motor assistance, representing an economic burden. To overcome these problems, a diversity of fall prevention strategies implemented on assistive devices such as smart walkers, have been widely explored. This study presents a novel strategy by using exclusively information from wearable sensors to detect near-falls while the subject uses a conventional rollator. A comparative analysis was performed to identify the most suitable classifier and the most relevant subset of features for detecting near-fall events. Ten able-bodied subjects performed 240 trials and simulated 180 near-falls with the rollator. The Ensemble Learning with the first 51 ranked features by the mRMR presented the best performance results (Accuracy = 95.18%; Detection time before recovery= 1.48±0.68 s). The results show that this strategy is suitable for use with conventional rollators, which are more used than smart walkers.