Abstract: Automated detection of epileptic seizures from EEG recordings is critical for patient monitoring and early intervention. We propose a hybrid Convolutional Neural Network (CNN)–Long Short-Term Memory (LSTM) architecture that ingests eight bipolar EEG channels (C3–P3, P3–O1, P4–O2, P7–O1, P7–T7, T8–P8-0, T8–P8-1, and FP1–F7) to detect seizure events. On the CHB-MIT scalp-EEG dataset (49 999 samples), our model achieves 97.92 % test accuracy, 93.29 % precision,
85.33 % recall, 89.14 % F1-score, and an AUC of 0.9807. Cross-validation yields comparable metrics. We also deliver an interactive Streamlit web app for real-time inference.
Index Terms: EEG, seizure detection, deep learning, CNN, LSTM, CHB-MIT dataset, Streamlit
|
DOI:
10.17148/IJARCCE.2025.14545
