LR-CNN based Classification for Improved EEG Epilepsy Seizure Detection using Discrete Wavelet Transform Technique

Epileptic seizures occur due to the functionality of neurological disorders in the brain, which can affect a patient's health. Machine Learning (ML) techniques are used for predicting epileptic seizures from the given dataset of Electroencephalograms (EEG) signals. However, the two major concerns are noise removal from pre-processing of EEG signals and feature extraction. To increase the computational speed of the preprocessing method, the epileptic seizure classification through a Discrete Wavelet Transform (DWT) is proposed here. The multi-resolution analysis for orthogonal scaling defines DWT with the same intimate of Daubechies wavelet used for analyzing the time-consumption of EEG signals. The Daubechies wavelet is used as a scaling function and implements the feature threshold extracted to determine the noisy data from the EEG signal. The CNN and LR were combined and utilized for the classification of EEG signals to detect epileptic seizures with increased sensitivity, specificity and accuracy of 91%, 89% and 97.2% respectively. The comparisons between the established classifiers were mainly based on exploring the receiver operating characteristic (ROC) curves and a sum of scalar performance measures concerning the classification