Development of Machine Learning Models to Detect Dynamic Disturbances in Human Gait

Machine learning has transformed the medical field by automating tasks and achieving objectives that are closer to human cognitive capabilities. Gait is a series of intricate interactions for humans, and identifying impaired gait is critical for effective decision-making in clinical practice. However, analysing the complex biological system that governs gait can be challenging. This research proposes a novel Criticality Analysis (CA) methodology to extract dynamic interactions in human gait and represent multivariate data in a nonlinear space. The proposed methodology characterises each data sample with a unique orbit, resulting from perturbations of a critical system composed of nonlinear controlled oscillators. The scale-free network of orbits is a quantitative measure of non-scale-free interacting sets of patterns, which reveal organised features of the structure of dynamic properties interconnected with human gait. This thesis focuses on implementing robust machine learning algorithms for effective detection and classification of complex dynamic patterns in human gait. The CA method maps gait features into a nonlinear representation, which is then used for training and testing categorisation algorithms. The proposed models utilise the Kernel property of the Support Vector Machines (SVM) classifier to identify high-order interactions between multiple gait data variables that may be challenging for traditional statistics. The algorithm was applied to three real datasets, and the SVM models designed using the CA method achieved an accuracy of 88.27% on average, compared to the K-Nearest Neighbours (KNN) approach's accuracy of 67.7%. The proposed SVM models use the receiver operating characteristics (ROC) and the area under the ROC metrics to evaluate their overall performance. The results of this research suggest that the proposed SVM models, with the support of the CA method, can perform as a robust and reliable classification tool for detecting dynamic disturbances of biological data patterns. This provides tremendous opportunities for clinical diagnosis and rehabilitation.

Grantor dc:publisher

Oxford Brookes University

Year dc:date

2022

Author dc:creator

• Eltanani, Shadi

Contributors dc:contributor

• olde Scheper, Tjeerd

Statement dc:rights

• All rights reserved

Language dc:language

en

DOI dc:identifier

https://doi.org/10.24384/ey4n-h547

OAI identifier oai:identifier

tle:cfafe796-968e-492e-a7a3-6a68e81de836:d6bd9758-527a-46cd-bfe2-c433766e8fca:1