Daubechies dalgacıkları kullanılarak EEG sinyallerinden epileptik nöbetlerin belirlenmesi

A seizure is a sudden and abnormal activity of brain, caused by disorderly electrical discharge of cerebral neurons. One of the most known seizures is recurrent attacks namely epileptic seizures which result from uncontrolled discharges of nerve cells. A traditional electroencephalogram (EEG) is often helpful to detect this uncontrolled process that indicates epilepsy. The visual detection of the seizures is required noticeable effort and time, especially in the long recordings, therefore alternative methods are needed. In the literature, there have been various epilepsy detection studies using different approaches. However, most of them utilizes patient-specific classifiers. In this study, wavelet-based algorithm is adopted to different classification methods for detection of epileptic seizures. To detect the effective frequency intervals of epileptic seizure, it is important to reach subbands of signal. Thus, choosing the suitable mother wavelet that resembles epileptic seizure is challenging. Consequently, ideal order Daubechies (db) wavelet becomes a critical point to achieve best performances. Here, it is selected Daubechies wavelets from db2 to db10 which have strong correlation with seizures to decompose the original EEG signal. The algorithm is applied on two different datasets. Eventually, the most appropriate eight features are used for classification model. Various machine learning algorithms namely, Decision Trees, Discriminant Analysis, Naive Bayes, Support Vector Machine (SVM), and k-Nearest Neighbor are used to differentiate epileptic and non-epileptic groups. The results show that db10 gives the best ACC performance of 95.83-100% in beta band using SVM classifier. This method can be a good alternative to the traditional approaches.