Parkinson's disease is a neurological condition that primarily impairs movement and motor function and significantly drops a patient's quality of life. Detecting PD early is necessary for effective disease management and timely medical intervention. In this research, we propose a machine learning system using vocal features as a non-invasive and accessible biomarker. Our study begins with exploratory data analysis, followed by feature selection methods to find the most impactful attributes. Then, we utilized several machine learning algorithms, including Decision Tree, Naïve Bayes, Support Vector Classifier, Random Forest, Logistic Regression,XGBoost, K-Nearest Neighbor, and AdaBoost, and calculated predictive performance using relevant metrics. To enhance our model performance, a hyperparameter tuning strategy is implemented to optimize the learning period. The XGBoost classifier, optimized through the principal component analysis feature selection technique and subsequent hyperparameter tuning, demonstrated robust performance. This work attained the highest accuracy of 97.44 % and shows that this fine-tuned model is both reliable and well-suited for the detection.