Abstract
Epilepsy, a common neurological disorder, can be monitored through electroencephalography signals, but current diagnostic methods largely rely on clinicians’ visual examination of these signals, making them susceptible to subjective interpretation. Recently, computer-assisted classification methods have been widely explored as more objective approaches for epilepsy diagnosis. The current study proposes a tree-based computer-assisted framework using theta band power features extracted from 35 EEG electrodes, including 19 unipolar and 16 bipolar channels. Three different tree-based classifiers were trained separately under three scenarios (unipolar electrodes only, bipolar electrodes only, and all electrodes without polarity distinction). For each scenario, the most successful classifier was used to rank the electrodes by importance and determine the most informative subset for epilepsy detection. The proposed method achieved high performance, with 86.8% accuracy, 84.3% F1 score, 82.3% sensitivity, and 90.7% specificity.