AN AUTOMATED PARKINSON’S DISEASE DIAGNOSIS: A FEATURE SELECTION APPROACH

Parkinson’s disease is one of the neurodegenerative disorders that significantly affect human health. Patients experience various negative effects such as tremors, walking disorders, and impaired speech. The disease also causes instability in walking, leading to tremors, and affects their writing skills. Studies on detection of disease generally focus on speech analysis. However, PD can be diagnosed by exploiting the loss of motor ability. In this work, a data set which was recorded at Cerrahpasa Faculty of Medicine, Istanbul University is considered. The data were collected from 15 healthy subjects and 75 with Parkinson’s Disease.by a graphic tablet. Each subject asked to draw a spiral in two different conditions which are named as static spiral test (SST) and dynamic spiral test (DST) respectively, and the drawings transformed into X, Y and Z axis of movement, Grip Angle, and Pressure data. During the study, the effectiveness of SST and DST conditions are considered. Various machine learning algorithms have been tested to determine the best classifier. The effect of features was also considered by utilizing a feature elimination process. As a result, the best classification performance was obtained as 93.55% by using Kernel Naïve Bayes network with SST data, by omitting Z axis.