Multi-class Epileptic EEG Classification Using a Compact Convolutional Neural Network for Clinical Brain–Computer Interfaces
Year: 2026
Type: conference-paper
Abstract: Epilepsy affects approximately 50 million people worldwide, with manual EEG analysis remaining time-consuming and error-prone. Automated seizure detection is essential for timely clinical intervention. This study presents a deep learning-based brain–computer interface (BCI) system for multi-class epileptic EEG classification, that goes beyond conventional binary approaches. In this work, NeuroSentry, a convolutional neural network (CNN) with three convolutional blocks, batch normalization, max-pooling, and dense layers (854,917 parameters) is implemented. The model is trained in 11,500 EEG samples from 500 subjects in five clinically relevant classes: seizure activity, tumor region, healthy tissue in tumor patients, and normal controls(eyes closed/open). Using the Adam optimizer (learning rate = 0.0001) and an 80:10:10 train–validation–test split, the model achieved 94.0% overall accuracy with 99% seizure detection precision and 99% recall, yielding a perfect ROC–AUC of 1.00. Per-class F1-scores ranged from 0.90 to 0.99, with only a 1% false-negative rate for seizures, indicating clinical robustness for real-time monitoring. The proposed multi-class framework offers enhanced clinical utility compared to standard binary classification models. The exceptional performance and balanced multi-class distinction demonstrate the effectiveness of the CNN for comprehensive EEG analysis and highlight its strong potential for clinical deployment.
Authors: Bikash Bora, Bikash Nath, Daisy Das, Nabamita Deb
Institution: Gauhati University
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7203627445