Efficient Parkinson’s disease classification from speech with filter-based feature selection and Genetic Algorithm–Bayesian Optimization ensemble integration

Parkinson’s disease (PD) is a progressive neurodegenerative disorder of the central nervous system that significantly impairs quality of life. Early and accurate diagnosis is essential to improve treatment outcomes and slow disease progression. In this study, we propose a computationally efficient and clinically feasible framework for PD classification based solely on vocal biomarkers. Our method leverages selective feature optimization and lightweight ensemble learning, avoiding reliance on deep or computationally intensive architectures. Three ensemble strategies are evaluated: (i) iterative majority voting, (ii) Genetic Algorithm (GA)-based classifier selection, and (iii) Bayesian Optimization (BO)-based probabilistic weighting. Building on these, we introduce a hybrid GA–BO ensemble method that combines GA-driven model selection with BO-guided weighting to optimize diagnostic performance. Experimental results demonstrate that the hybrid ensemble achieves state-of-the-art metrics, including 96.4% accuracy, 97.6% F1-score, and a Matthews Correlation Coefficient (MCC) of 0.906. The proposed system is adaptable across diverse feature sets and suitable for integration into mobile or edge-computing platforms. Overall, the framework offers a non-invasive, scalable, and cost-effective decision support tool for early-stage PD diagnosis.