Abstract:
Brain-computer interface (BCI) is a system used as a pathway between Human brains and machines to communicate without any physical interaction. Feature Selection is one of the core concepts in BCI, which hugely impacts the model's performance. Feature Selection is the process where you automatically or manually select those features which contribute most to your prediction variable or output in which you are interested. But the main problem for BCI is identifying the related features from a set of data and removing the irrelevant or less important features that do not contribute much to our target variable to achieve better accuracy for our model. We aim to make a genetic algorithm that can extract optimal features for a hybrid-based BCI. This study highly focusses on optimal genetic algorithm for features election of hybrid EEG-EOG, EEG-EMG, and EEG-fNIRS based brain-computer interface (BCI). Recent studies have not revealed the concept of the genetic algorithm used in efficient exploring feature space. The designed genetic algorithm has explored the low fitness value of those features and the whole feature space by using SVM as an objective function. Several hybrid modalities are being tested with different number of channels along with different window sizes combination sets i.e. 2-, 3-, 4-. While utilizing this genetic algorithm on the EEG-EMG dataset, an increase of 4% accuracy has been observed for the 4-features combination set. In contrast to EEG-FNIRS, accuracy is improved by 5% with the window size of 5sec. Whereas in EEG-EOG, the reported accuracy was 67% and we achieved maximum average accuracy i.e 78% for the 4-feature combination set. The reported results depict the improved accuracy with the designed genetic algorithm. In addition, with the use of explored list and several different checks, the problem of redundant features is resolved.
