Jumanto Unjung, Rofik, Sumarni Adi, Annisaa Utami, Much Aziz Muslim
Parkinson's Disease (PD) is a progressive neurodegenerative disorder that requires accurate early diagnosis to support effective clinical intervention. Machine learning approaches using voice-based features have shown potential for non-invasive PD detection; however, high-dimensional feature spaces and class imbalance remain significant challenges. This paper proposes an optimized classification framework that integrates Principal Component Analysis (PCA), the Synthetic Minority Over-sampling Technique (SMOTE), and Extreme Gradient Boosting (XGBoost), with hyperparameter optimization using GridSearchCV. A comparative analysis is conducted to evaluate the impact of dimensionality reduction by examining PCA-based and non-PCA-based feature representations. Experimental results demonstrate that the PCA-SMOTE-XGBoost model outperforms the non-PCA configuration, achieving an accuracy of 94.87% compared to 92.31%, along with improved precision, recall, and F1-score. Furthermore, PCA reduces feature dimensionality from 22 to 8 components and decreases training time, indicating improved computational efficiency. These results demonstrate that the proposed framework effectively enhances Parkinson's disease classification performance while maintaining robustness and efficiency. © 2026 IEEE.
Universitas Negeri Semarang, Faculty of Mathematics and Natural Sciences, Department of Computer Science, Semarang, Indonesia; Universitas AMIKOM Yogyakarta, Faculty of Computer Science, Yogyakarta, Indonesia; Telkom University, Department of Informatics, Purwokerto, Indonesia
