LoginAbstract
Motor development during childhood is closely linked to cognitive, linguistic and social development. Because locomotor control emerges early in life, quantitative gait assessment can provide indicators for the early identification of developmental impairments. However, investigating motor development remains challenging due to the multidimensional nature of locomotor control and the difficulty of interpreting large sets of biomechanical parameters. Recent advances in wearable inertial sensing technologies enable the acquisition of gait data in real-world scenarios and the extraction of both temporal and nonlinear descriptors of human movement. Despite the availability of these metrics, identifying which aspects of locomotor behaviour best characterize motor development remains an open problem. This work investigates whether clustering techniques applied to gait features extracted from inertial measurement units (IMUs) can reveal distinct locomotor control strategies in school-aged children. A dataset of 442 participants performing Natural Walking, Tandem Walking and Place-and-Bricks tasks was analysed. From the inertial signals, temporal and nonlinear gait parameters describing variability and movement complexity were extracted. A preliminary statistical assessment of the features was performed to verify dataset consistency. Unsupervised clustering algorithms were then applied to identify groups of subjects exhibiting similar locomotor patterns. To maintain interpretability and avoid the typical black-box behaviour of machine learning models, explainability techniques were used to determine which features most strongly contributed to the resulting clusters. The proposed framework provides an interpretable approach for exploring locomotor development patterns in childhood and highlights how combinations of temporal variability and nonlinear gait descriptors may reveal different motor control strategies.
Altri metadati
Gestione del documento:
