Connectivity Measures based Imagined Speech Recognition using parallel ReliefF-NCA based Channel Selection

Recently imagined speech-based BCI systems are becoming quite popular among researchers because they are not dependent on external stimuli, and their generation is easy. Even though so much work is done related to imagined speech, they are still not feasible. In order to make practical imagined speech-based systems that are cost-efficient and easy to use, systems having higher accuracies, even with a small number of electrodes, are required. In an imagined speech, one electrode to 64 electrodes EEG devices were used, and it is seen that accuracy decreases when the number of channels is reduced. As the specific region for imagined speech production is still debatable, location-specific electrode distribution is not suitable in all scenarios. Therefore, the need for a suitable channel selection method arises that can reduce the number of channels without affecting accuracy. This work proposes a parallel ReliefF-NCA-based channel selection method in which, firstly parallel combination of ReliefF and Neighborhood Component Analysis (NCA) optimized using a Bayesian algorithm is used for feature selection. Furthermore, with the help of the importance of features related to a particular channel, the selection of channels is performed.