Feature Extraction from Mental Arithmetic Based EEG Signals

Analysis of mental arithmetic based electroencephalography (EEG) signal can be helpful for patients who have difficulty learning or understanding arithmetic or have autism spectrum disorders. It is difficult to separate mental arithmetic from EEG signals since these signals are nonstatic and nonlinear. In this study, we extracted features based entropy, skewness and entropy + skewness from the EEG signal. Then, extracted features were classified by support vector machines. The average 85.69% classification accuracy (CA) was calculated from the entropy based features that best determine the mental arithmetic of the EEG signals. This value is 9.79% higher than the average 75.90% CA calculated in the literature. This result indicates that proposed method is effective for this data set.