A hybrid scheme using PCA and ICA based statistical feature for epileptic seizure recognition from EEG signal

Abstract

Epilepsy is commonly regarded as a neurological disorder which can be characterized by repetitive unprovoked seizures. Electroencephalogram (EEG) is the neuro-physiological measurement of the brain's electrical activity recorded by electrodes placed in the cerebral cortex. The EEG signals play an essential role in the diagnosis of epilepsy. This paper proposes an approach for classifying the epileptic seizure patterns that carry significant indications regarding chronic neurological disorders. In this regard, a dimensionality reduction scheme, hybrid in nature, utilizing Independent and Principal Component Analysis (ICA and PCA) is developed followed by the extraction of statistical features for epileptic seizure identification. At first, a particular number of sub-frames is extracted from the given EEG signal. The extracted sub-frames are considered as the input of PCA and ICA. After that, the significant components of ICA and PCA are utilized for statistical feature extraction. Lastly, the supervised support vector machine (SVM) classifier is employed for the seizure classification purpose. To evaluate the raised method, the publicly available EPILEPTIC dataset is used. According to the experiments and result analysis, the proposed scheme achieves convincing performance in terms of accuracy when the first components of ICA and PCA algorithms are used for feature extraction.