An interpretable and balanced machine learning framework for Parkinson's disease prediction using feature engineering...

Area of Science:

• Neurology
• Computer Science
• Biomedical Engineering

Background:

• Parkinson's disease (PD) is a progressive neurological disorder with diagnostic challenges.
• Machine learning (ML) offers potential for precise and efficient PD prediction.
• Early and accurate diagnosis of PD is crucial for patient management.

Purpose of the Study:

• To develop an enhanced ML framework for improved PD prediction.
• To integrate data balancing, feature selection, and explainable AI (XAI) techniques.
• To enhance the fairness, performance, and interpretability of PD diagnostic models.

Main Methods:

• Evaluation of nine ML algorithms on clinical and voice features.
• Application of Synthetic Minority Oversampling Technique (SMOTE) and NearMiss for class imbalance.
• Utilized Featurewiz, Tree based Feature Importance, and chi-square for feature selection.
• Employed SHAP and LIME for XAI to interpret model decisions.

Main Results:

• The KNN model with SMOTE achieved 92% accuracy, 0.94 F1-score, and 0.95 G-Mean, indicating balanced and reliable PD detection.
• Some models showed higher accuracy (up to 97%) on imbalanced data but lacked sensitivity and balance.
• Feature selection identified key voice biomarkers like Pitch Period Entropy (PPE) and Noise-to-Harmonic Ratio (NHR).

Conclusions:

• Combining SMOTE, feature engineering, and XAI significantly improves ML model fairness, performance, and interpretability for PD prediction.
• The proposed framework provides an accurate and interpretable ML-based diagnostic tool.
• This research supports early PD diagnosis and enhances patient management strategies.