Title: Machine learning comparative study for human posture classification using wearable sensors

Authors: Aaron Rasheed Rababaah

Addresses: College of Engineering and Applied Sciences, American University of Kuwait (AUK), Salmiya, 30012, State of Kuwait

Abstract: Human posture classification plays important role in number of applications including elderly monitoring, workplace ergonomics, sleeping patterns studies, sports, fall detection, etc. Despite of the fact that the topic is well-studied in the literature, many studies utilise one to few models to investigate the classification reliability of different postures. In this paper we present a rich study of the problem with six primary machine learning algorithms and an overall of nine different models considered in training and testing the real world collected data of human subjects. In this study, six different postures are addressed namely: sleeping, sitting, standing, running, forward bending and backward bending. The study considered two categories of models, supervised and unsupervised learning algorithms. After intensive training and testing of all algorithms, multi-layer perceptron and K-Means outperformed other algorithms with an impressive classification accuracy of 99.88%.

Keywords: human posture; wearable sensors; machine learning; neural network; MLP; multi-layer perceptron; nearest neighbour classification; discriminant analysis; self-organising maps; K-means; GMM; Gaussian mixture model; clustering; classification; signal processing.

DOI: 10.1504/IJCSM.2023.133518

International Journal of Computing Science and Mathematics, 2023 Vol.18 No.1, pp.54 - 69

Received: 07 Dec 2020
Accepted: 15 Apr 2021
Published online: 19 Sep 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article