Index of Physical activity and Fall Efficacy scale classification through biomechanical signals and Machine Learning.

Abstract

The rapid increase of the elderly population and chronic diseases have augmented disability in today's world. This situation has led researchers and engineers to create tools and technologies that allow health caregivers, physical trainers, and health policymakers to understand, measure, and treat people with disabilities. Nowadays, artificial intelligence techniques have been applied to improve the performance of these technologies. This article shows the development of a novel classifier that utilizes Machine Learning (ML) algorithms and biomechanical signals to predict a subject's International Physical Activity Questionnaire (IPAQ) and Falls Efficacy Scale (FES). Three ML algorithms were applied K-Nearest Neighbors (KNN), Decision tree, and Support Vector Machine (SVM). Results show the accuracy of classification over 95%, 99%, and 89%, respectively, and validate the correlation between qualitative scales and biomechanical responses in balance training. This classifier poses as an innovative tool to help professionals adjust and improve physical training programs.

A Deep-Learning Based Real-Time Prediction of Seated Postural Limits and Its Application in Trunk Rehabilitation

“…Furthermore, this gate-controlled regression problem needs a hierarchical combination of two types of tasks, so parallelly combining two output branches [21] was not a good choice. Another possible solution was to simplify the problem to a classification problem (i.e., segmenting subjects' workspace into hierarchical ellipses [22]), but ellipses cannot accurately represent the sitting workspace. We solved this problem using statistical learning (i.e., picking a mixture probabilistic model to describe the model's target space and designing the loss based on the maximum likelihood principle) and achieved priority tuning of tasks and gatecontrolled training by adding weights to the two meta terms of the proposed loss.…”

A Deep-Learning Based Real-Time Prediction of Seated Postural Limits and its Application in Trunk Rehabilitation

“…Furthermore, this gate-controlled regression problem needs a hierarchical combination of two types of tasks, so parallelly combining two output branches [21] was not a good choice. Another possible solution was to simplify the problem to a classification problem (i.e., segmenting subjects' workspace into hierarchical ellipses [22]), but ellipses cannot accurately represent the sitting workspace. We solved this problem using statistical learning (i.e., picking a mixture probabilistic model to describe the model's target space and designing the loss based on the maximum likelihood principle) and achieved priority tuning of tasks and gate-controlled training by adding weights to the two meta terms of the proposed loss.…”