<![CDATA[Early detection of Parkinson's disease (PD) is essential to enhance patient quality of life through timely intervention. This research aims to develop a predictive model using an ensemble learning approach and optimal feature selection. This experimental study employs three machine learning algorithms: random forest, XGBoost, and extra trees, optimized through hyperparameter tuning, feature selection techniques, and Kernel Principal Component Analysis (KPCA) for dimensionality reduction. The study utilizes the UCI Machine Learning Parkinson Dataset, which consists of 80 samples and 44 acoustic features extracted from patients' voices as they sustain the vowel sound "/a/" for five seconds. Results show that XGBoost achieved the highest accuracy at 88.93% after tuning and KPCA, followed by extra trees with 86.15%, and random forest with 85.47%. The application of KPCA successfully reduced data dimensionality without sacrificing accuracy, thereby improving modeling efficiency. These findings suggest that voice data holds significant potential for early PD detection and that selecting appropriate algorithms and dimensionality reduction techniques is crucial for optimizing data-driven diagnostic models.]]>
Copyrights © 2024
