SHAP-based interpretable machine learning for Parkinson's disease severity prediction: integrated analysis of clinical and environmental features

Introduction Parkinson's Disease (PD) represents the second most prevalent neurodegenerative disorder globally, with traditional assessment methods suffering from limitations including substantial inter-rater variability and inability to capture multifactorial complexity underlying disease progression. Methods Based on data from 500 Parkinson's disease patients, we integrated 7 standardized clinical phenotypes (excluding UPDRS to prevent data leakage) and 8 environmental exposure factors, compared 10 machine learning algorithms using 5-fold cross-validation, and applied SHAP interpretability analysis for transparent feature importance assessment. Results XGBoost with SMOTE sampling achieved clinically meaningful discriminative performance (AUC = 0.781, precision = 0.548, recall = 0.750) appropriate for screening applications. SHAP analysis revealed non-motor symptoms as the primary predictor (SHAP value = 2.76), followed by serum dopamine concentration (2.39) and age (2.16), while environmental factors demonstrated modest but statistically significant contributions. Discussion This proof-of-concept study developed an interpretable framework with methodological safeguards against data leakage, demonstrating promising screening potential with realistic performance expectations. However, the cross-sectional, single-center design limits generalizability, requiring external validation and longitudinal studies before clinical deployment.