Logistic regression Gaussian mixture model and softmax discriminant classifier for epilepsy classification from EEG signals

Abstract

The patient suffers from spontaneous seizures under a neurological condition termed epilepsy. Due to the severe disturbances happening in the electrical activity of the brain, the seizures occur. An epileptic seizure is simply a transient occurrence of symptoms because of excessive and abnormal neuronal firing of neurons in the brain. Due to the hypersynchronous and abnormal activity of neurons in the brain, epilepsy manifests itself in the form of seizures. The most disabling aspect and the characteristic feature of this disease are witnessed by the occurrence of seizures. It is quite difficult to identify the presence of epileptic activities in order to characterize and understand the spatio-temporal patterns. The automatic classification and detection procedure of seizures from Electroencephalography (EEG) signals is quite crucial for the localization and epileptic seizure activity classification. The seizures are typically non stationary and highly dynamic in nature so that the rhythmic activities cannot be easily interpreted. In this paper, a two level classification procedure is employed, firstly the risk of epilepsy is classified with Softmax Discriminant Classifier (SDC) and secondly it is further optimized and classified with Logistic Regression Gaussian Mixture Model (LRGMM) Classifier. Results show that a classification accuracy of about 93.38% is obtained when classified with SDC and a classification accuracy of about 97.91% is obtained with classified with LRGMM.