Applying machine learning to track and analyze human movement

Abstract

Traditional statistical methods comparing movement kinematics in biomechanics involvecalculating discrete variables and comparing between pre-defined groups based on age, sex, pathology, or condition. However, this method generalizes that the movement patterns of individuals within a group are similar and inherently different from those of the group being compared to, which might not be the case. Clustering analysis, which takes an unlabeled set of data and organizes them into homogenous groups, may provide a unique way to distinguish movement patterns within a group of individuals. However, there are many things to consider when clustering any dataset including each clustering algorithm and evaluation measure taking different approaches, leading to different clustering results. This dissertation proposed a method of determining the most appropriate clustering model by comparing k-means and hierarchical clustering (HCA) on different types of biomechanics time-series data using several evaluation measures. Using a majority ranking method, we were able to generate movement profiles of lumbar and pelvis angles during a trunk flexion/extension, head and trunk anterior-posterior (AP) acceleration during steady-state gait, and pelvis AP linear acceleration and vertical (V) angular velocity during a timed up and go (TUG) test. Overall, it was found that most clusters generated on each dataset did not contain a single age or sex demographic. The differences observed between movement profiles were the magnitude and timing of the clustered variable. Clustering analysis was able to separate participants that may have compromised ability to control the head during steady state walking and were slower at turning during the turn-around subphase of the TUG test. Overall, our findings underscore the utility of clustering analysis in elucidating subtle variations in movement behavior in a wide range of different movement evaluations, thereby enhancing our understanding of functional mobility and falls risk assessment in diverse populations.