Predicting Parkinson's Disease Progression: A Non-Invasive Method Leveraging Voice Inputs

Parkinson's Disease (PD) is a complex neurodegenerative condition with a global impact, demanding precise disease progression prediction to facilitate effective treatment strategies. To assess PD symptoms, the Unified Parkinson's Disease Rating Scale (UPDRS) is widely adopted, encompassing both motor and non-motor assessments. This research delves into voice inputs as a non-intrusive method to predict total UPDRS and motor UPDRS scores, offering new possibilities for Parkinson's assessment. Feature engineering and data augmentation techniques address challenges related to class imbalance and diverse demographics, including an original imbalanced dataset with more females than males. Additionally, three new datasets are created: oversampled balanced, only-female, and only-male datasets. Ensemble-based stacking model, including random forest and extreme gradient boosting as base models and the gradient boosting regressor as the meta-regressor, demonstrate promising performance and robustness in predicting UPDRS scores, showcasing the efficacy of voice inputs for PD assessment. Furthermore, the feature importance analysis provides insights into crucial contributors influencing predictions. Various performance metrics, such as accuracy, mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and R-squared (R2), are used to evaluate the model’s performance. Additionally, by incorporating telemonitoring capabilities, the voice-based approach offers the possibility of remote and continuous PD assessment, allowing for real-time monitoring and early detection. This advancement could significantly improve the quality of life for PD patients and facilitate more personalized and effective treatment plans.