Wavelet-Enhanced CNN-BiLSTM Model for Epileptic State Classification
Nadjla Bettayeb Amina Achaibou Nidaa elislam Benferdia
AbstractEpilepsy is a neurological disorder that affects approximately 1% of the global population, and its accurate detection is crucial for effective diagnosis and management. Electroencephalography (EEG) is a widely used, non-invasive tool for monitoring brain activity in epileptic patients. This paper presents a deep learning-based framework for classifying epileptic EEG signals using a hybrid Convolutional Neural Network (CNN) - Bidirectional Long Short-Term Memory (BiLSTM) architecture. The model processes raw EEG signals as well as time-frequency decompositions using the Discrete Wavelet Transform (DWT), enabling better representation of non-stationary brain dynamics. To assess performance, we conducted extensive binary and multiclass classification experiments on the Bonn EEG dataset. Binary tasks included distinguishing epileptic vs. non-epileptic activity, seizure prediction (interictal vs. ictal), and healthy brain states under different conditions. Our model, especially when trained with DWT features, achieved a maximum accuracy of 99%, with 100% sensitivity and 98% specificity, outperforming raw data. Multiclass classification results also showed strong generalization, with the DWT-enhanced model reaching 89.5% accuracy when classifying among healthy, interictal, and ictal states. These results confirm that combining time-frequency decomposition with a CNN-BiLSTM architecture offers a robust solution for reliable and accurate EEG-based epilepsy detection.