Explainable Voting Classifier Reveals Non-Physiological Signal-Derived Features In Eye Movements Associated With Parkinson’s Disease

Parkinson’s disease (PD) is a neurodegenerative disorder characterized by motor and non-motor symptoms such as oculomotor dysfunction. This study implements a comprehensive Machine Learning pipeline to process eye-tracking data as continuous time series of PD patients and healthy controls. This study deviate from traditional event-based analyses by employing the Time Series Feature Extraction Library (TSFEL) to derive a rich set of statistical and spectral features alongside traditional kinematic parameters. Thirteen ML-based classifiers were evaluated, and the four top performing algorithms, LinearSVC,Ridge Classifier, K-Nearest Neighbors, and Logistic Regression, were integrated into a Voting Classifier. The Voting model achieved classification performance of 90.25% in accuracy, precision, recall, and F1-score. SHapley Additive exPlanations (SHAP) and Accumulated Local Effects (ALE) plots were seamlessly integrated into the pipeline for interpretability. These Explainable AI (XAI) tools revealed that the model's decision making process is predominantly influenced by signal-based, non-physiological features, such as Pupil_0_Absolute energy and Pupil_0_Spectral spread. By validating the importance of these novel features against established PD pathophysiology, this study delivers a transparent, high-performance, and explainable AI description, effectively bridging the gap between advanced data modeling and clinical applicability.