Classification for Imperfect EEG Epileptic Seizure in IoT applications: A Comparative Study

Epileptic seizure detection could be detected through investigating the electroencephalography (EEG), which is deemed to be very important for IoT wearable sensor-based health systems. EEG-based classification is crucial for a wide-range of applications to analyze real-time vital signs using features concerning predefined set of data classes. The aim of this paper is to conduct a comparative study for several classification techniques and demonstrate the effect of uncertainty in the EEG data on the classification accuracy. We define a model for decomposing the EEG using various transformation such as discrete cosine transform, discrete wavelet transform into several sub-bands. After feature extraction, a comparative study to assess the classification algorithms' performance is conducted. In addition, we evaluate their overall accuracy and complexity as performance measures. For this purpose, we use the support vector machine (SVM) and the Artificial Neural Network (ANN). These are chosen as classifier models to study the performance of the obtained features. The discussion will include the evaluation of the classifiers' performance using the EEG-based epileptic seizure data in two categories, noiseless and noisy. In addition, there are some statistical features extracted to characterize the complete EEG data feeding to these two classifiers. A publically available EEG dataset is employed for both normal and epileptic seizure for automatic epileptic seizure detection as a benchmark.