A new feature extraction and classification mechanisms For EEG signal processing

The Electroencephalogram (EEG) signal processing is one of the extensively used research field in recent days, in which the epileptic seizure detection and classification plays an essential role. It is very hard to determine the network of EEG signal, because it contains enormous and fluctuated information about the actions of the brain. Therefore, the earlier research works focused to detect the seizure based on its features. For this reason, it implemented various feature extraction, feature selection, and classification algorithms during the abnormality prediction. But, it has the major drawbacks of inefficient classification results, utilizing large amount of features, and increased complexity. To solve these problems, this paper aims to develop a Masking and Check-in based Feature Extraction Technique (MCFET) and an integrated K-Means with K-Nearest Neighbor classification algorithms for detecting whether the EEG signal is normal or abnormal. Initially, the features of the signal such as Signal to Noise Ratio, variance, and Standard Deviation are extracted by generating the mask value based on the check-in function. After that, the extracted features are given as the input for classification, where both the Euclidean distance and Chebyshev distance are computed. Based on the minimum similarity value, the signal is classified as normal, or abnormal (i.e. interictal or ictal). In experiments, the performance of the proposed MCFET with modified k-NN are evaluated by using different performance measures. Also, the superiority of the technique is proved by comparing it with some existing techniques.