Analysis and classification of EEG signals using a hybrid clustering technique

This paper presents a novel hybrid approach based on clustering technique (CT) and least square support vector machine (LS-SVM) denoted as CT-LS-SVM for classifying two-class EEG signals. The study aims to extract representative features from the original EEG data through the CT method and then to classify two-class EEG signals by the LS-SVM using these features as inputs. In order to test the effectiveness of the proposed method, the experiment is carried out on an epileptic EEG data and a mental imagery tasks EEG data. The classification accuracy of the current method is compared to the previous reported methods of the literature. The proposed approach is found to achieve an average classification accuracy of 99.19% for the mental imagery tasks EEG data and 94.18% for the epileptic EEG data. Our results show the highest classification accuracy (99.90%) for healthy subjects with eyes open (Set A) and epileptic patients during seizure activity (Set E) from the epileptic EEG data among the reported algorithms. Thus, the findings of the current research demonstrate that the CT method is efficient for extracting features representing the EEG signals and the LS-SVM classifier has the inherent ability to solve a pattern recognition task for these features.