A Robust Deep Learning Model to Predict Epileptic Seizures based on Electroencephalographic (EEG) Signals

Abstract

Epileptic seizures are sudden, uncontrolled electrical disturbances in the brain, which can lead to significant medical complications if not accurately diagnosed and treated. Electroencephalography (EEG) is widely used for monitoring brain activity, offering a non-invasive method to detect seizures. This paper presents a novel approach Deep Neural Optimum Transformation (DNOT) for epileptic seizure prediction based on EEG signals using a hybrid deep learning model, Convolutional LSTM AutoEncoder (CLSTM-AE). The proposed model leverages the strengths of Convolutional Neural Networks (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) layers for temporal dependency modeling. The AutoEncoder component ensures dimensionality reduction and reconstruction, allowing the model to focus on the most relevant features of EEG signals. The DNOT mechanism optimizes the feature extraction and learning process, transforming EEG data into an optimal latent space for improved classification performance. We evaluated the model using the Epileptic Seizure Recognition dataset, comparing its performance against nine existing machine learning and deep learning models. The proposed model achieved a remarkable accuracy of 96.58%, significantly outperforming conventional models like Random Forest, XGBoost, and deep learning models like ANN and LSTM. Other performance metrics, including precision (95.43%), recall (96.12%), and Fl score (95.77<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>), further demonstrate the model's effectiveness. This methodology offers a reliable, high-accuracy solution for seizure prediction, providing significant potential for improving medical diagnostics.

Topics & Concepts