A Hybrid Wrnn and Dbn-Based Approach (Hymod) For Parkinson’s Disease Detection Using Voice Data

Abstract

Parkinson's Disease (PD) is a progressive neurodegenerative disorder that requires early and accurate detection for effective treatment. Existing approaches for PD detection have utilized deep learning models, incorporating techniques like Synthetic Minority Oversampling Technique (SMOTE) for data balancing and min-max normalization for feature scaling. However, min-max normalization can be sensitive to outliers, potentially skewing the model's performance. Additionally, traditional classifiers may struggle with feature selection, leading to suboptimal results and increased risk of overfitting. The current models, while effective, face challenges with generalization and the accurate detection of PD, particularly when working with sequential voice data, where temporal dynamics are crucial for diagnosis. This study suggests an approach for addressing these problems improved hybrid model (HYMOD) combining a Weighted Recurrent Neural Network (WRNN) and Deep Belief Network (DBN). The proposed method applies SMOTE for data balancing, but replaces min-max normalization with Z-Score normalization to mitigate the impact of outliers and ensure more stable model convergence. An entropy-based butterfly optimization the feature selection process uses an algorithm, improving model efficiency and focusing on the most relevant features, reducing noise and redundant data. By leveraging the sequential processing capability of WRNN and the deep feature extraction of DBN, the hybrid model significantly outperforms existing methods, achieving superior accuracy, precision, recall, and F1-scores for early PD detection. This enhanced model, through its innovative integration of advanced pre-processing, feature selection, and classification techniques, offers a more robust solution for reliable PD diagnosis and timely intervention..