Abstract: Epileptic seizure is associated with significant morbidity diseases and mortality. An early identification of seizure activity can help prevent patients from adverse outcomes. Electroencephalography (EEG) raw data is a good source to recognize epileptic seizure from other brain activities. Numerous previous studied have applied feature engineering techniques to extract clinical meaningful features in order to indentify Seizure from EEG raw data. However, these techniques required intensive clinical, radiology and engineering expertise. In this study, we applied 6 machine learning algorithms (including naïve bayes, logistic regression, support vector machine, random forest and K-nearest neighbours and gradient boosting decision trees) and 3 deep learning architecture (including convolutional neural network (CNN), long-short term network (LSTM) and Gated Recurrent Unit (GRU)) to conduct binary and multi-label brain activities classification. Our best results of binary classification yielded that ensemble classifiers can classify seizure from other activities with a high accuracy and AUC over 0.96. In multi-label classification, both GRU and RNN yielded an averaged accuracy over 0.7. A compared study was also presented to analyze the performance of each configuration. In conclusion, machine learning and deep learning demonstrated their potential usage in epileptic seizure identification using EEG raw data. Future work may be experimented in a larger dataset to enable the seizure identification at a timely manner.Abstract: Epileptic seizure is associated with significant morbidity diseases and mortality. An early identification of seizure activity can help prevent patients from adverse outcomes. Electroencephalography (EEG) raw data is a good source to recognize epileptic seizure from other brain activities. Numerous previous studied have applied feature engineering ...Show More
