Machine Learning-Based Detection of Epileptic Seizures from EEG Signals for Scalable Clinical Screening

This paper presents a lightweight, scalable epileptic seizure detection system using non-invasive electroencephalogram (EEG) signals. We applied and compared several supervised machine learning models, such as Logistic Regression, Support Vector Machines (SVM), and Random Forests (RF), on a publicly available EEG dataset from the University of Bonn. We applied feature engineering techniques such as statistical descriptor extraction and z-score normalization to enhance model performance and applicability. Our model with the best results, the Random Forest classifier, had an accuracy of 98.3% and a ROC-AUC of 0.995, outperforming all baselines. Analysis of feature importance revealed signal variance, entropy, and peak-to-peak amplitude as the most predictive biomarkers for seizure detection. We further confirmed the usefulness of these features through model interpretability methods. Notably, the entire framework relies on computationally inexpensive methods, and thus it may be possible to integrate it into mobile or embedded EEG diagnostic systems for application in resource-scarce and underserved clinical settings. The work highlights the promise of simple yet powerful machine learning pipelines for real-time seizure monitoring and screening.