Machine learning based detection of compensatory balance responses to lateral perturbation using wearable sensors

Loss of balance is prevalent in the older population and also in people who have mobility impairment. The primary aim of the present paper is to develop an efficient classifier to automatically distinguish compensatory balance responses (or near-falls) from regular stepping patterns. In this study, 5 young, healthy subjects were perturbed by lateral pushes while walking and the compensatory reactions were recorded by three wearable inertial measurement units (IMUs). Time domain features of these signals were extracted and reduced, using different dimension reduction methods, i.e., PCA, SPCA and KSPCA. The performance of k-nearest neighbor (k-NN) and support vector machines (SVMs) classification methods for detection of compensatory balance responses is investigated. The results of this study advances wearable measurement methods to accurately and reliably monitor gait balance control behavior in at-home settings (unsupervised conditions), over long periods of time (i.e., weeks, months). Building on the current study, subsequent research will examine ambulatory data to identify balance recovery processes for clinical assessment of fall risk.