A Comparative Study Between Classical Feature Engineering and RNNs for Seizure Detection in Imbalanced Data

Epilepsy is one of the most common neurological diseases worldwide, defined as a number of seizures in the brain that affect the person's quality of life and her/his ability to perform regular activities. Epilepsy is diagnosed in several ways, one of which is a test known as EEG. The process of examining brain activity is a long and error-prone process. Researchers developed machine learning algorithms to identify epileptic seizures through available datasets. This study aims to compare the performance of a Machine Learning ML model with classical feature engineering approach and RNN for seizure detection in imbalanced dataset. In the experiment, feature engineering improved the recall dramatically in the classical pipeline and slightly improved it in RNN where there is no (manual) feature engineering., The AUC also had increased after feature engineering in classical pipeline and RNN, exceeding 95%. it was clear that the feature engineering improved the performance dramatically in the classical pipeline, and that the order of the feature engineering and resampling (downsampling) didn't affect the performance, and although applying a classical ML model on manually extracted features and RNN gave approximately close results, the recall which spots every seizure was better in RNN comparing to the classical approach.