Exploring foot pressure patterns for skiing experience classification in a simulator-based environment

Abstract This study investigated whether deep learning can distinguish groups of ski experience based on plantar-pressure distributions measured during simulator-based training and whether classifier outputs change during repeated training. Weight-shift images from 27 participants were classified into beginner, novice, and intermediate groups, operationally defined by self-reported skiing exposure, using a YOLO11-based image classifier. Participant-wise generalization was explored using leave-one-participant-out (LOPO) cross-validation, with performance summarized at the participant and session levels. The classifier showed moderate discrimination between the beginner and novice groups but limited generalization to the intermediate group. Session-level classification achieved an accuracy of 0.667, macro-F1 of 0.489, and balanced accuracy of 0.524. Longitudinal changes in classifier outputs were subsequently examined in six beginner participants using fixed participant-dependent models trained only on baseline data. Mixed-effects analyses provided tentative evidence of nonlinear changes in classifier outputs across repeated training sessions, although the results were sensitive to model specification, and substantial inter-individual variability was observed. Qualitative Eigen-CAM visualization and biomechanical analyses further indicated that the classifier responded to distributed plantar-pressure characteristics. These findings support the feasibility of using plantar-pressure patterns to characterize skiing-experience groups and to track longitudinal changes in simulator-derived pressure patterns, while highlighting the need for larger sample sizes and independently validated skiing-skill criteria before such models can be used for objective proficiency assessment.