International Journal of Computer Applications |
Foundation of Computer Science (FCS), NY, USA |
Volume 57 - Number 16 |
Year of Publication: 2012 |
Authors: Hussein El Saadi, Ahmed Farouk Al-sadek, Mohamed Waleed Fakhr |
10.5120/9202-3733 |
Hussein El Saadi, Ahmed Farouk Al-sadek, Mohamed Waleed Fakhr . Informed Under-Sampling for Enhancing Patient Specific Epileptic Seizure Detection. International Journal of Computer Applications. 57, 16 ( November 2012), 41-46. DOI=10.5120/9202-3733
Thirty percent of epileptic patients encounter intractable seizures, (seizures that do not respond to medication), thus, an accurate seizure detector would help improve their quality of life. Unfortunately, seizure detection is one of the many fields that suffer from imbalanced dataset i. e. the ratio between ictal and inter-ictal records is huge which makes it difficult to build an accurate classifier. This paper attempts to build a classifier that is able to overcome the previously mentioned challenge by dividing the dataset in ensembles and utilizing multiple SVM classifiers. As a result, the detector was able to reach an overall accuracy of 97. 3%; thus, opening the field for building strong classifiers from highly imbalanced datasets in the biomedical domain.