Improving Parkinson's Disease Diagnosis with Machine Learning Methods

Abstract

Parkinson's disease is a type of disease caused by the loss of dopamine-producing cells in the brain. As the amount of dopamine decreases, the symptoms of Parkinson's disease emerge. Parkinson's disease is a slow-developing disease, and symptoms such as hands, arms, legs, chin and face tremors are increasing over time. As the disease progresses, people may have difficulty in walking and speaking. There is no definitive treatment for Parkinson's disease; however, with the help of some drugs, the symptoms of the disease can be reduced. Although there is no definitive treatment for Parkinson's disease, the patient can continue his normal life by controlling the problems caused by the disease. At this point, it is important to prevent early detection and progression of the disease. In this study, different types of classification methods such as Logistic regression, Support Vector Machine, Extra Trees, Gradient Boosting and Random Forest are compared in order to predict Parkinson's disease. A total of 1208 speech data sets consisting of 26 features obtained from Parkinson's patients and non-patients were used in the classification stage. The feature space of the dataset is expanded due to correlation maps. These correlation maps are constructed with the features which are obtained by using Principal Component Analysis (PCA), Information Gain (IG) and all features respectively. It is concluded that, classification results which are attained with expanded features outperform the classification results attained with the original features of the data.