Brain seizures detection using machine learning classifiers based on electroencephalography signals: a comparative study

The paper demonstrates various machine learning classifiers, they have been used for detecting epileptic seizures quickly and accurately through electroencephalography (EEG), in real time. Symptoms of epilepsy are caused abnormal brain activity. Analyzing and detecting epileptic seizures presents many challenges because EEG signals are non-stationary, and the patterns of the seizure vary for each patient. Moreover, the EEG signals are noisy, and this affect the process of seizure detection. On the other hand, Machine learning algorithms are very accurate, adaptive and generalize very well when provided with diverse and big training data and can easily analyze complex structure of the EEG signal despite the noisiness when compared to other methods. With this approach the features of epileptic seizures can be learned and used to correctly identify other seizure cases. The demonstration states a comparison between various classifiers, including random forests, K-nearest neighbors (K-NN), decision trees, support vector machine (SVM), logistic regression and naïve bayes. Different performance metrics is used such as accuracy, receiver operating characteristics (ROC), mean absolute error (MAE), root-mean-square error (RMSE) and most importantly detection time for each algorithm. The Bonn university dataset has been used for demonstration process for the classification of the epileptic seizure.