This work aims to develop a brain–computer interface (BCI) system based on electroencephalogram (EEG) signals, that is capable of remote controlling rehabilitation systems using wireless connections. This system can extract delta waves from raw EEG in real-time to predict motor imagery (MI) tasks. Where we built a simple acquisition device that acquires EEG signals using three dry electrodes, these non-invasive channels are positioned on the scalp surface at the occipital and central lobes. After the acquisition step, we amplify the signals and remove permanent noise during the preprocessing step. Then, in the feature extraction step, we extract possible features from each channel. Then, we select only some important features at the feature selection step, by the calculation of each feature’s contribution score. In the classification phase using machine learning algorithms, we select the light gradient boosting machine (LGBM) algorithm enhanced by the multi-verse optimization (MVO) algorithm, which enables the building of optimum prediction models. Also, this work employed a data analysis phase. Where to evaluate the characteristics independent between features at each step, we analysed the data using the correlation matrix results. As well as, we analysed the data changes temporally and spatially between MI tasks at each step. Therefore, the classification results indicated that the system accuracy score is over 90%. While in related work, we have an accuracy value ranging between 79% and 89%. These comparative results show the best quality of our system proposed for this work-based delta wave.
Read full abstract