A Machine Learning Approach for Parkinson's Disease Prediction Using Voice Analysis: A Study on Logistic Regression, AdaBoost, and Principal Component Regression

Parkinson's Disease (PD) is a movement and speech neurodegenerative disorder that gets progressively worse with time, for which early detection is essential for improved patient prognosis. Machine learning methods have indicated potential in diagnosing PD with voice-based characteristics. In the present work, we used and compared three machine learning models, namely Logistic Regression (LR), AdaBoost, and Principal Component Regression (PCR), to predict PD from a dataset of both PD and normal voice recordings. The dataset was preprocessed, involving data cleaning, feature scaling, and train-test splitting, to achieve maximum model performance. Models were trained and tested with appropriate statistical voice features like jitter, shimmer, fundamental frequency, and noise-to-harmonics ratio. The performance was measured by accuracy, precision, recall, and F1-score. Logistic Regression proved to be the best model with an accuracy of 92.47%, followed by AdaBoostwith an accuracy of 91.02% and Principal Component Regression with achieving accuracy of 88.45%. The findings indicate that logistic regression, a simple and easy-to-interpret model, is very effective in separating PD from healthy subjects on the basis of voice features. Although ensemble learning methods such as AdaBoost and dimensionality reduction methods such as PCR were also effective, their accuracy was slightly less. These results indicate the promise of voice analysis with machine learning for early and non-invasive detection of PD. Future research can investigate how deep learning methods and larger datasets can be used to increase predictive accuracy and robustness.