This study presents ESDRCX, a multimodal ensemble combining decision trees, SVM, random forest, a CNN for spiral images, and an XGBoost meta-learner optimized with Optuna, reporting ~95.7% accuracy on the HandPD dataset for Parkinson's detection.

Improves diagnostic accuracy and early detection workflows, but provides no mechanistic, biomarker, or therapeutic insights relevant to drug discovery, limiting its translational value for PD therapeutics.

Parkinson's disease (PD), a progressive neurological disorder affecting motor function, has been significantly rising in prevalence in recent years. Current diagnostic methods, relying on clinical observations, neurological exams, and periodical DaTscan imaging, may exhibit reduced sensitivity in the early stages. To develop a robust and multimodal machine learning model for early detection, an Ensemble Approach (ESDRCX) is proposed that integrates a meta-ensemble stacking technique that incorporates Decision Tree, Support Vector Machine (SVM) and Random Forest using quantitative data, along with a Convolutional Neural Network (CNN) for spiral image input. Additionally, the outputs are merged using XGBoost as the meta-learner optimized with Optuna-based Tree-structured Parzen Estimator (TPE). The ESDRCX attains a prominent 95.7% accuracy, 86% precision, 91% recall, 88.6% F1-score and 87% AUC with the HandPD dataset, denoting a significant progress in Parkinson's disease diagnostics. The proposed framework delivers an accurate, interpretable and computationally effective approach for early PD detection.