Classification of performance in risk-of-falls assessment based on accelerometer data and feature boosting

Abstract

Clinical assessments of risk of falls for the elderly are growing in relevance. This work presents a strategy for the evaluation of the collective capabilities of features and indexes, collected with accelerometers, in discriminating good and poor responses in the first two stages of the Dynamic Gait Index assessment of risk. Adaptive-boosting machine learning and cross-validation strategies are implemented to test the overall classification capabilities of these indexes, their consistency and coherence with the biomechanical expectations for these tests. As a result, select indexes of performance are highlighted, and the relevance of these features and parameters to performance in these parts of the DGI assessment, as consistent, quantitative measures of performance, supplementary to evaluations by specialized physicians, is tested.