Improved LSTM-SqueezeNet Model and Improved Kalman Filtering Process for Epileptic Seizure Prediction Using EEG Signal

Epilepsy is a neurological condition resulting in irregular activity in the brain’s neurons, which leads to frequent unexpected seizure occurrences. Hence, it is essential to predict this disorder at an early stage that prevents the patient’s life. Various Deep Learning (DL)-based techniques are employed in this aspect. Still, the models rely on inaccurate performance with improper training. In this work, a hybrid classification system consisting of SqueezeNet and Improved Long Short-Term Memory (ILSTM) models is proposed to predict epileptic seizures using Electroencephalography (EEG) inputs. The prediction of seizure involves three stages. The EEG signal initially undergoes preprocessing, where the improved Kalman filter is employed to preprocess the signal. After that, several features such as Empirical Mode Decomposition (EMD), improved Stockwell transform, spectral density, spectral skewness, and statistical features are derived from the preprocessed signal. The extracted features are subjected as the input to the prediction stage with a hybrid classification system utilized. The hybrid system is the integration of ILSTM and SqueezeNet classifiers. Additionally, the performance of the proposed method is compared to that of conventional models using a range of performance metrics. As a result, the suggested model shows superior performance compared to the existing techniques.