Machine Learning Approach for Classification of Parkinson's Disease Using Vertical Ground Reaction Force Gait Biomarkers
Parkinson's disease (PD) is conventionally assessed by neurologists who rate clinical manifestations on the Unified Parkinson's Disease Rating Scale (UPDRS). Such rating depends on examiner expertise is subjective and is difficult to scale. This paper presents a supervised machine-learning gait-classification system that assists clinicians in detecting PD and estimating its severity stage. Gait is a sensitive biomarker of motor impairment accordingly the framework operates on the vertical ground reaction force (VGRF) gait dataset and derives a minimal feature vector through statistical analysis of spatial (force-based) and temporal (stride, stance, swing) gait descriptors. The Shapiro-Wilk test validates the distributional assumptions of each feature and correlation-based feature selection (CFS) identifies the salient mutually non-redundant biomarkers. Four supervised classifiers (decision tree (DT), support vector machine (SVM), ensemble classifier (EC), and Bayes classifier (BC)) are then trained for statistical and kinematic analyses that predict disease status and severity. Evaluation follows a subject-independent nested cross-validation protocol using accuracy, precision, recall, specificity, F1-score, and AUC, and the results are compared against recent published methods. Quantitative results accuracy is 97.4% and F1-Score 97.3%.
Authors
- B Hariprasad (ORCID: https://orcid.org/0000-0002-3509-900X)
- Hari Nadha Reddy Y N
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.5281/zenodo.24365845
- Primary Topic
- Balance, Gait, and Falls Prevention
- Type
- article
- Field-Weighted Citation Impact
- 0.00