Classification of electroencephalography signal using statistical features and regression classifier

Enormous digital electroencephalography (EEG) acquisition systems available nowadays for researchers due to the high demand in the brain signal research. Using EEG-based emotion recognition, the computer can look inside a user head to observe their mental state of sad and happy emotion. Thus, there is a need for efficient mechanism to detect those emotions accurately along with computation complexity. The current algorithms available are excessively complex with higher computational time. In this study, 14 channels of EEG signals acquired from emotive device with 128 Hz sample rate. These raw signals undergo preprocess stage using band pass and ICA filter. This research focuses two components which is feature extraction and classification. A combination of statistical features has been carrying out to extract important signal. To classify the EEG signal into sad and happy classes, Support Vector Machine (SVM) and Linear Regression has been applied. Waikato Environment for Knowledge Analysis (WEKA) as training tools is employ to train the dataset and test the accuracy of the classifier. Results presented that Linear Regression has better detection accuracy with 95% compared to SVM with 80% average accuracy. In conclusion this research suggests using Linear Regression for future work on predicting between sad and happy emotion from the EEG signal