EEG based over-complete rational dilation wavelet transform coupled with autoregressive for motor imagery classification

Motor Imagery (MI) based Brain-Computer Interface (BCI) applications are designed to analyse how the brain interacts with the external environment from electroencephalograph (EEG) signals. Despite current models achieving promising results, developing an accurate classification of MI from EEG signals remains a significant challenge. In this paper, we designed an MI classification model named (ORDWT_AR) utilising an over-complete rational dilation wavelet transform (ORDWT) coupled with an autoregressive (AR) model. Firstly, EEG recordings are segmented into intervals using a sliding window method. Then, each EEG segment is passed through the ORDWT to analyse EEG signals. As a result, a series of stop bands is obtained from each segment. Then, the AR is adopted and integrated with ORDWT to extract representative features from each EEG interval. The selected features are sent into several classification models, including Weighted k-nearest Neighbour (WKNN), Decision Tree (DTree) and Boosted Trees (BST). Four benchmark EEG databases were used to evaluate the proposed model, three of which were collected from brain-computer interface (BCI) Competition III and one from the CHB-MIT. The results demonstrated that the proposed model ORDWT_AR coupled with the WKNN classifier achieved an average of 99.8% classification accuracy for the three BCI competition III datasets and 99.7% for the CHB-MIT dataset. The obtained results revealed that the proposed scheme is a promising tool for classifying EEG signals and has outstanding results. The proposed model can support experts in aiding disabled people to interact with their environment accurately and improve the quality of their lives.