Classification of Seizure Types Based on Statistical Variants and Machine Learning
Anand Shankar, Samarendra Dandapat, Shovan Barma
Abstract
The majority of the research works are successfully applying advanced machine learning algorithms to classify epileptic seizures using electroencephalograms (EEG). Certainly, the accurate classification of epileptic seizure types can play a significant role in the prognosis and treatment of epileptic patients’ conditions. In this work, machine learning classifiers — artificial neural network, decision tree, k–nearest neighbor, random forest, and eXtreme boosting gradient have been employed to classify complex partial seizure, focal non-specific seizure, generalized non-specific seizure types, and seizure-free. For this purpose, statistical variants — mean, skewness, kurtosis, standard deviation, approximate entropy, and energy have been extracted from EEG segments. Thenceforth, machine learning algorithms performed multi-class epileptic seizure type classification based on these variants. Furthermore, using the principal components analysis methodology, the classification of epileptic seizure types has been analyzed using the lower dimensions of statistical variants sets. For evaluation of the proposed method, a publically available EEG dataset contributed by the Temple university hospital (TUH, v1.5.2) has been taken into consideration. The classification accuracy of multi-class epileptic seizure types has achieved up to 100%. The experimental performances demonstrated that the proposed work can efficiently and accurately classify the seizure types.