Machine Learning Techniques for Voice-based Early Detection of Parkinson's Disease

Neurodegenerative, progressive diseases having multiple motor and non-motor characteristics are mainly associated with Parkinson's disease (PD). A person suffering from PD usually faces problems with his/her speech or vocal impairment in the early stage of this disease. Presently, one of the significant research fields is biomedical signal processing. Henceforth, considering vocal impairment as early signs for PD patients, diagnosis systems using vocal analysis is the need of the hour. Early detection of Parkinson's disease may aid in better diagnosis and treatment of the disease, as well as better equipping caregivers to care for the patient. Further, this may also help the hospital management centers better use their resources. This paper aims to discuss and analyze the various machine learning methods for predicting the early onset of PD. The UCI Machine Learning repository has the dataset we used in our research. Various machine learning methods are applied to this dataset, and the performance of each method is further explored. For the proposed model, we observe that stacking different learning models together works best for the given task with an accuracy of 93%.