QFS-MSGRU: Quantum Feature Selection Approach Integrate With Multi-Scale GRU Networks for Robust Parkinson’s Disease Prediction

Parkinson’s disease (PD) is a neurodegenerative disorder, and early diagnosis is a critical challenge due to the complexity of symptoms and heterogeneity of speech-based biomarkers. Traditional machine learning (ML) models are not always effective in dealing with high-dimensional feature spaces, causing poor quality classification accuracy. Overcoming such challenges, we proposed the QFS-MSGRU Model, a deep learning framework that combines Quantum-inspired Feature Selection (QFS) and Multi-Scale Gated Recurrent Units (MSGRU) with self-attention to facilitate PD prediction. The method starts with data preprocessing stages of feature normalization and class balancing with SMOTE. A quantum-inspired feature selection technique employs the most discriminative features while maintaining fundamental information without a significant loss of dimension. The classification model employs a multi-scale GRU to learn temporal relationships at multiple time scales, and a self-attention mechanism facilitates long-range feature interactions. The experimental results show that QFS-MSGRU achieves 96.46% accuracy, 97.30% precision, 95.58% recall, F1-Score 96.43% and an AUC score of 99.06%, significantly outperforming baseline techniques. This work introduces a novel and efficient Deep Learning (DL) model for PD prediction, showing enhanced classification performance through quantum inspired feature selection, multi-scale recurrent modeling, and attention mechanisms. The results confirm its real-world potential for clinical applications, facilitating improved early diagnosis and automated screening systems for neurodegenerative diseases.