Speech-based Parkinson’s Disease Detection with Hybrid Machine Learning Approach

Parkinson’s disease (PD) is a prominent neurodegenerative disorder that affects millions of people worldwide. It leads to irregularities in motor function, speech, and cognitive abilities. Early detection of PD is crucial, as addressing the disease in its initial stages can result in better management and slower progression. This study explores the application of ensemble machine learning (ML) methods to enhance the accuracy and reliability of early PD detection. The ensemble techniques—Bagging, Boosting, Voting, and Stacking—work by combining the strengths of multiple models to overcome the limitations of traditional approaches, thereby improving diagnostic performance. These methods were applied to a feature set of 11 features that includes vocal biomarkers (MDVP, jitter, shimmer, and HNR) etc, which reflect the effects of PD on voice. The combination of ensemble learning as a strong diagnostic tool provides an overall accuracy of 97%, which was achieved in our experiment using 195 vocal samples (147 PD and 48 healthy). The high accuracy levels attained here seem to point toward ensemble methods, potentially providing a clearer advance for PD detection over single deep learning models. In the future, these methods can be validated on larger and more heterogeneous datasets and methods from deep learning will be explored to refine the diagnostic model even further.