Machine-learning classification of postural sway in young adults during colored noisy vestibular stimulation

Stochastic resonance suggests that adding an optimal level of noise can enhance a weak signal, making it detectable. This report presents a secondary analysis of data from Gavriilidou et al. (2025) examining how noisy galvanic vestibular stimulation (nGVS) affects postural control. A k-nearest neighbor (KNN) classifier was used to distinguish center-of-pressure (CoP) trajectories recorded from healthy young adults standing on a firm surface with feet together and eyes closed. CoP data were analyzed using seven time-domain variables and 84 time-frequency bandwidths in the forward-backward and side-to-side directions. Three time-domain and two time-frequency features were selected for classification. Model accuracy was evaluated for differentiating among stimulus intensities (% perceptual threshold), noise types (Pink or White), and responsiveness to the perturbation. Classification accuracy exceeded 96% for all conditions, indicating distinct CoP patterns. The model further distinguished participants who did or did not exhibit a stochastic-resonance response to nGVS. SHapley Additive exPlanation analysis revealed that feature contributions were greater under White-noise stimulation. These findings demonstrate that nGVS systematically modulates postural control and that machine learning can effectively capture its condition-specific influence on balance dynamics.