Electroencephalogram (EEG) is the main diagnostic tool for the monitoring, diagnosis and treatment of epilepsy which is a neurological disorder. EEG signals can disrupt easily by involuntary movements that are called artifact contaminants such as blinking, coughing. In this study, the artifacts in the extrac- and intracranial EEG signals have been cancelled out from the EEG with the use of Kaiser window based Finite Impulse Response (FIR) filter. The most important features in the EEG signals have been selected by the Principle Component Analysis (PCA) method. The selected features have been classified by applying ensemble learning methods that are Boosting, Bagging and Random Subspace. The aim of this study is to increase the extrac- and intracranial EEG signal classification by calculating window spectral parameters. The algorithms' classification performances have been compared in terms of accuracy rates, sensitivities, specificities, prediction rates and training times according to the 5 x 5 cross validation. Subspace KNN algorithm, as revealed by results, is higher than the other algorithms' classification performances.