Parkinson’s Disease (PD) is a disorder of the central nervous system which affects movement, often including tremors. Nerve cell damage in the brain causes dopamine levels to drop which gradually degrades the functionality of the brain. Since PD is a neurodegenerative ailment, Electroencephalography (EEG) signal are used for early detection of Parkinson’s Disease. EEG being non-linear and non-stationary manual analysis is not only time consuming but prone to error. To detect PD, two methods are discussed in this paper: (1) CNN for EEG images and (2) k-nearest neighbors for manually extracted features from EEG signals. The proposed methodology is applied to publicly available datasets (1) University of New Mexico (UNM) (27 PD patients and 27 controls) and (2) Iowa (14 PD patients and 14 controls). Data from New Mexico is used to evaluate the performance of the model using k-fold cross-validation method and data from Iowa is used for out-of-sample evaluation. Mean test accuracy on the mentioned datasets reaches to 88.51% and 87.6% respectively making an improvement of 3.11% and 1.9% for UNM and Iowa dataset, as compared to the current state-of-the-art accuracy.