An Explainable Machine Learning Framework for Parkinson’s Disease Classification Using High-Dimensional Speech Features

Parkinson’s disease is a progressive neurodegenerative disorder in which early and reliable diagnosis remains clinically important. Speech impairment is one of the most common manifestations of the disease and offers a non-invasive basis for automated screening. This study proposes an explainable machine learning framework for Parkinson’s disease classification using high-dimensional speech features. The framework integrates data preprocessing, feature selection, multiple machine learning classifiers, probability-based evaluation, and explainability analysis to build an interpretable and robust diagnostic model. Four classifiers were applied: Logistic Regression, SVM with an RBF kernel, Random Forest, and XGBoost. Experimental evaluation on the speech-feature dataset showed that XGBoost achieved the best overall classification performance, while Random Forest produced the most reliable calibration. In addition, feature-importance analysis revealed that dynamic cepstral and TQWT-based speech descriptors were among the most influential predictors for Parkinson’s disease detection. Oversampling strategies, including Random Oversampling, SMOTE, and ADASYN, did not improve performance over the original data distribution, indicating that model-level robustness was more beneficial than synthetic class balancing for this dataset. Overall, the findings demonstrate that explainable ensemble machine learning provides an effective and interpretable approach for Parkinson’s disease classification from high-dimensional speech biomarkers.