ABSTRACT Parkinson's Disease (PD) is a chronic neurodegenerative condition characterized by loss of dopaminergic neurons in a specific region of the brain. Symptoms such as hand tremors, walking difficulties, and impaired communication become noticeable in individuals with PD. Given this problem, early and accurate detection of PD remains a key change in clinical practice. The objective of this research is to design a robust and explainable framework for PD detection based on speech signals analysis. A hybrid 1D CNN—BiLSTM framework was designed to capture spatial feature patterns and temporal dependencies. Recursive Feature Elimination (RFE) was applied to select 13 most discriminative speech features, while Synthetic Minority Oversampling Technique (SMOTE) was integrated with 5‐fold cross‐validation to address class imbalance. Ablation studies assessed the contribution of each model component, and confusion matrix analysis enabled clinical interpretation by quantifying true positives, true negatives, false positives, and false negatives. The experimental findings demonstrated that the proposed 1D CNN—BiLSTM framework achieved strong predictive performance of 92.10% accuracy, 96.43% precision, 93.10% recall, 94.33% F1 score, and clinical reliability with high true positives indicating reliable patient identification and few false negatives reducing risks of missed diagnoses when compared to alternative models. In conclusion, the proposed model demonstrates novelty by integrating explainability, feature selection, and robust validation. Its application provides a non‐invasive and reliable framework to support Parkinson's disease screening and early clinical decision making.
Olawuyi et al. (Wed,) studied this question.