Optimized Electroencephalogram Preprocessing with Discrete Wavelet Transform and EfficientNet for High-Accuracy Seizure Detection
Abstract
Abnormal neuronal activity in the brain leads to seizures which seriously affect quality of life and can become life-threatening in certain situations. Automatic seizure detection research matters because timely intervention requires early accurate detection. Deep learning achieves high effectiveness in seizure detection while its success relies strongly on how well EEG signals undergo pre-processing. The accuracy and reliability of models require effective pre-processing of raw EEG data because it contains noise and patient variability. The research introduces a deep learning model dedicated to seizure detection while emphasizing advanced pre-processing and flexible adaptability. The model transforms EEG signals into a two-dimensional feature representation to enhance spatial and temporal pattern recognition capabilities. The model's modular structure provides adaptability across various hardware platforms which facilitates both optimization processes and custom solutions. The model reached a 93.27% classification accuracy on two seizure datasets when utilizing EfficientNet with DWT showing its powerful generalizability and practical effectiveness in real-world situations.