Integration of Continuous Wavelet Transform and Convolutional Neural Network for Multiclass EEG Dataset Classification

Abstract

This thesis presents a novel method for EEG classification in time-frequency domain using deep learning architecture. Existing deep learning architectures suffer from poor performance when classifying EEG data in Time-frequency domain. The proposed method seeks to improve classification process and provide better accuracy and loss than previously has been achieved. The Continuous Wavelet Transform is used to convert brainwaves into time-frequency domain and then Convolutional Neural Network is used for feature learning and classification of EEG data. The results have been cross-validated by Kfold cross validation and Leave-One-Out Cross Validation(LOOCV). The proposed method has also been compared with VGG16, Google Net, AlexNet models. This model produces results on publicly available dataset: Epilepsy dataset by UCI(Machine Learning Repository)