Answers are grounded in this paper and cite their sources.
The artificial intelligence introduction to signal processing is constrained by the scarcity of labelled data, particularly for gait analysis which has a great impact in bio-measurement and progressive neurological conditions such as Parkinson's disease. Synthetic data augmentation, in which artificial training samples are generated to expand the effective dataset, offers a route around this constraint. This study develops a novel gait multi-conditional Wasserstein generative adversarial network (G2AN ) with gradient penalty to generate synthetic vertical ground reaction force gait signals conditioned on Hoehn & Yahr severity stage and evaluate them as augmentation for downstream Parkinson's disease classification. Synthetic-data augmentation raised binary Parkinson's disease detection accuracy from 65% to 85% and four-class severity staging accuracy from 32% to 76%, with the largest gains observed on the most under-represented severity class. The approach suggests that generative augmentation may offer a tractable route to improving severity classification on extremely small gait datasets.
No takes yet. Share an insight, caveat, or question.
Gosney et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: