Comparative Study for Classification Methods to Predict and Detect Epilepsy Seizure

Epilepsy is a medical problem that tackles lots of patients. It limits the life activity of such patients due to the seizures that occur anytime and anywhere. Thus, creating a monitoring system that could make their life easier and allow them to perform daily life activities as safe as possible has been studied. This paper is a continuity for a project where a system monitors epileptic patients using a set of wearable sensors connected to several processing units, the main sensors are Electroencephalogram (EEG) for the electrical activity of the brain, electrocardiogram (ECG) for the electrical activity of the heart, and accelerometers to monitor the posture of the patient, features extracted from EEG and ECG signals are used for prediction/detection of seizures. Thus, the objective of this paper is to present a comparative study in order to choose the appropriate classification method for seizure prediction and detection. Five classification methods were studied: Decision Trees, Discriminant Analysis, Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Ensemble Learning. For each of the classification methods, the accuracy, the specificity, the sensitivity, the precision, the false detection rate and the Fmeasure were studied and compared. The results for seizure prediction system show that the highest accuracy, specificity and F-Measure, and lowest false detection rate were achieved with Bagged Trees Ensemble Learning, the highest sensitivity and precision were achieved with Quadratic Support Vector Machine. However, for seizure detection system, the highest accuracy and F-Measure were achieved with Linear Support Vector, the highest sensitivity and specificity were achieved with Fine Gaussian Support Vector Machine, and the highest precision and lowest false detection rate were achieved with Subspace Discriminant Ensemble Learning.