Classification of EEG Recording in an Epileptic Patient Using DWT and Extreme Gradient Boosting Classifier
The study of electroencephalogram (EEG) signals for the determination of seizures involves several background interference and artifacts. This has become a great challenge for accurate detection of epileptic seizure signals with proper predictive analysis. The cognitive approaches for the prediction of seizures have been a booming challenge for the medical practitioners to exactly create a separation between the ictal state and non-ictal states from the long term EEG recordings. This paper gives an insight into the proper prognosis of epileptic seizures in a more automated way. Different classes of the EEG benchmark dataset i.e UBonn dataset are taken for the analysis. A binary classification model is designed along with 8 level Daubechies order 4 decomposition using DWT. The statistical features for each of the subbands are extracted. Then 70% train set and 30 % test set are split for classification purposes. Here a balanced data method called SMOTE method is implemented for making the training feature set balanced. Extreme Gradient Boosting classifier (XGBC) is implemented, which outperforms another state of art classifiers for different classes of data taking different performance metrics. This model for classifying EEG signals can better demonstrate the physical implementation of the unexpected occurrence of seizures by medical practitioners. It yields less complexity with increased performance on a reduced feature dimension.