I. INTRODUCTION

EEG (Electroencephalography) is an important and effective tool to measure the activities of brain and understand the complex behavior of the brain. Even a small variations in EEG signal define a specific type of brain disorder and hence it is necessary to develop an algorithm which can be used for the detection of brain abnormality since it may affect the life of a person. EEG is a dynamic non-invasive and relatively inexpensive method used to monitor the state of brain. According to the survey about 1% of the world’s population is having epilepsy and it is the 2nd most common neurological disorder [3]. This neurological disorder is associated with recurrent, unprovoked epileptic seizures which results due to the excessive – discharge of central neuron groups which are harmful to human. However, person who does not have epilepsy can suffer from seizures. The Behavior of the person with seizure ranges from simple finger twitching to convulsion where muscles will contract and relax in a manner which can not be controlled, results in an involuntary movement of the body. Automatic seizure detection can be divided into five groups: frequency domain based, time domain based, timefrequency domain based, nonlinear methods and artificial neural network based [4]. Since EEG is a non stationary signal, it is more appropriate to use time-frequency based method such as discrete wavelet transforms (DWT). DWT analysis is important to address the different behavior of EEG signal on both time and frequency domain [5]. Once the feature has been extracted, then the resultant data will be given to the classifier to classify signal as normal or abnormal. Many methodologies have been used for the classification purpose. In the proposed work neural network classifier has been used.

II. REVIEW

Many researchers are trying to propose different methodologies for the automatic detection of brain disorder using EEG signals. In a paper by M. Mursalin [1] an algorithm has been proposed for automated epileptic seizure detection using improved correlation-based feature selection with random forest classifier. The algorithm has tested on 5 set of EEG data. Each consists of 100 channels. Two set was recorded from a normal person and 3 were recorded from epileptic seizure patient. Feature extraction was done using time domain, frequency domain and entropy based. Best feature were selected using ICFS (Improved Correlation-based Feature Selection method) and classifier used was Random Forest classifier (RF). In another paper by J. G. Bogaarts [2] has worked on a method for the seizure detection in which the EEG data that is to be incorporated into the baseline buffer are automatically selected based on a novelty detection algorithm. For the classification purpose SVM classifier has been used. Zandi [2] has proposed a wavelet based technique for the real time detection of epileptic seizure. Decomposition of EEG signal was done using wavelet packet transform and frrequency band which represnets maximum separation was obtained. The algorithm was experimented on 14 pateints with 75.8 h with 63 seizures. The method resulted in 90.5% of sensitivity and false detection rate of 0.51/h. Yinxia Liu [6] has described an algorithm using wavelet transform and SVM for sizure detection. The data set they have used consist of 2 to 5 h of seizure data and 24 to 26 h of non-seizure data and. The decomposition of multi-channel iEEG (intracranial EEG) was done and frequency bands were extracted from that only three bands are selected. Features were extracted and then sent to SVM for classification purpose. The specificity achieved was 95.26% and seneitivity of 94.46% with false detection rate of 0.58/h.