Identification of whole-body reaching movement phenotypes in young and older active adults: an unsupervised machine learning approach

Abstract

Studies reported age-related motor control modifications in whole-body movement in several aspects of spatiotemporal movement organization by comparing young and older adults. However, studies on motor control involve high complexity and high-dimensional data of different natures, in which machine learning has proved to be effective. Furthermore, conventional studies focus on comparisons of movement parameters based on a priori grouping, whereas unsupervised machine learning allows the identification of inherent groupings within the dataset. The current investigation was carried out by using the unsupervised machine learning on motor control features across age-groups. An important question was whether we could identify different movement patterns based on motor control features and whether they were age-dependent or independent. We investigated motor control parameters variations in a whole-body reaching movement across young and active older adults including woman and man (n=19). We applied the K-means clustering algorithm to segment the kinematic data (21 features) of all individuals. We propose a methodology applying the latest recommendations for clustering methods in the field of whole-body movement motor control. Analysis revealed two distinct motor control patterns which were age independent. The first pattern exhibited higher shoulder, ankle and knee angular excursions, along with a higher vertical velocity of center of mass (CoM), compared to the second pattern, which had higher hip and back angular excursions, along with a lower vertical velocity CoM. The clustering methodology demonstrated its effectiveness to identify distinct motor patterns based solely on motor control features independently of age-grouping.Significance StatementK-means clustering algorithm enabled us to identify two distinct age-independent motor patterns: a first pattern with high shoulder, ankle and knee angular excursions, and vertical velocity of CoM; a second pattern with high hip and back angular excursions and low vertical velocity of CoM.Demonstrates how unsupervised machine learning can identify motor patterns and proposes a methodology to apply it in the field of whole-body movement motor control.Proves the complementary contribution of unsupervised machine learning to conventional approach for motor control studies, which enables to process the high complexity and dimensionality of movements.Advances understanding of motor behaviours through unsupervised machine learning analysis of whole-body reaching movements.