XGBoost-Based Classification of Epileptic versus Non-Epileptic EEG Signals

Abstract

Epilepsy is a chronic neurological disorder that affects more than 50 million people worldwide. The electroencephalogram (EEG) is the clinical gold standard for its assessment; however, manual inter pretation is time-intensive and subject to inter-observer variability. This work presents an automatic binary classification system for distinguishing epileptic from normal EEG signals using the Extreme Gradient Boosting (XGBoost) algorithm. The Epileptic Seizure Recognition dataset from the University Hospital of Bonn was used, comprising 11,500 one-second EEG segments, each described by 178 temporal amplitude samples, with a 4:1 class imbalance. The pipeline integrates exploratory analysis in the time and frequency domains, class-imbalance compensation through the scale pos weight parameter, and 5-fold stratified cross-validation. The XGBoost classifier achieved an AUC-ROC of0.9953 on the held-out test set and 0.9951± 0.0009 in cross-validation, with 97.57% accuracy and an F1-score of 0.9389 on the epileptic class. Under the same protocol, it obtains a discriminative ability statistically equivalent to that of Random Forest and Support Vector Machine, while providing the most balanced sensitivity–specificity trade-off and requiring 2.3× to 3.5× less training time. These results support XGBoost as an accurate and computationally efficient candidate for automated EEG screening in resource-limited clinical settings.