Comparison of WPD, DWT and DTCWT for Multi-Class Seizure Type Classification

Epilepsy is characterized by recurrent seizures that come in diverse types which are treated in a variety of ways. Electroencephalography (EEG) is a technique that is frequently used in medical settings to diagnose epileptic seizures. The accurate identification of seizures helps to provide optimal treatment and accurate information to the patient. This paper compares three wavelet-based feature extraction methods for multi-type seizure classification using EEGs. WPD, DWT, and DTCWT are used to extract features from EEG data, which are then classified using LightGBM. We evaluated the proposed methods on the TUH EEG Seizure Corpus v1.5.2, which is the world's largest available EEG epilepsy database. We also examined three different combinations of features on three different problems, each containing different seizure classes according to their medical definitions. The performance of our proposed method is measured according to the overall F1-score. For patient-wise cross-validation, EEG-based seizure type classification using WPD achieved the best results of weighted F1-score of 64%, 66.6% and 83.97% for 7-class, 5-class, and 2-class problems respectively. The results are compared with existing state-of-the-art techniques and our results established new benchmark results for this dataset.