Detection of Parkinson's Disease using Machine Learning with Feature Analysis from Audio Signals

Background: Early detection of Parkinson's Disease (PD) is crucial for timely intervention and improved patient outcomes. PD affects motor functions and often results in vocal impairments, making voice analysis a promising non-invasive diagnostic approach. Methods: This study evaluates the effectiveness of machine learning algorithms—Decision Tree, Random Forest, Logistic Regression, Support Vector Machine, Naive Bayes, K-Nearest Neighbors, and XGBoost—for PD detection using a dataset of vocal features from the UCI Machine Learning Repository. The dataset, comprising 23 vocal attributes, was preprocessed using the Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. Models were assessed using metrics such as Accuracy, F1-Score, Recall, Precision, and ROC-AUC. Results: The Random Forest and Support Vector Machine algorithms demonstrated the highest classification accuracy (96.6%) and superior overall performance. These findings highlight the potential of these methods for distinguishing between PD-positive and PD-negative cases using voice data. Concluding Remarks: This study underscores the potential of machine learning in developing accessible and accurate tools for early PD diagnosis. Further enhancements, such as feature refinement and dataset expansion, could improve these methods' generalizability and robustness, supporting their integration into healthcare systems.