Detection of Parkinson’s Disease with Machine Learning Support

Abstract

Parkinson’s disease (PD) is a neurodegenerative disorder causing partial or complete loss of motor reflexes and speech. It affects the patients significantly in their daily life. The accurate diagnosis of PD is quite complex, requires a lot of resources, and equipment and is time-consuming for diagnosis, especially in its initial stage’s occurrence of the disease. To help the medical experts and researchers diagnose PD, we developed a machine learning approach based on the simple descriptive tasks captured in audio on the smartphone. The dataset for demonstrations consists of over 1500 patients with approximately 9% of patients diagnosed with PD. Hence the imbalance of the dataset the evaluation metrics such as Mathews correlation coefficient (MCC), sensitivity - specificity, and ROC curve were picked to describe the selected machine learning algorithms performance. The analysis of tested results in a single approach resulting with 73.49% (52.53% - 91.41%) MCC.