Improving EEG Signal Classification with Recurrent Neural Networks for Brain-Computer Interfaces

Brain-computer Interfaces can change the way humans interact with tech and their surroundings. This brainsensing technology allows control devices to interface with machines through thoughts naturally and even endows the most common person with superhuman capabilities. BCIs typically use electroencephalography signals interpreted to translate brain activity into commands for such devices: the low signal-to-noise ratio and high inter-subject variation. In the following section, we will introduce a new approach, Recurrent Neural Networks, to classifying EEG signals and discuss how this method helps alleviate some of the challenges traditional methods face. RNNs are quite efficient for sequence prediction tasks as they can capture the relationship between data sequences. We combine RNNs into the classification stage to ensure EEG signals are classified quickly and accurately in real time, enhancing BCI performance. We pre-process by applying a filter bank to the EEG signals and extracting features using a CNN. An RNN is then applied over these features, which learns the dependencies in time for a given EOS event (output). We suggest a transfer learning model with the only motivation of mitigating the disadvantage of needing plenty of training data for this model and making it more practical in real-world applications. The preliminary results indicate that our proposed method improves the classification accuracies on EEG signals compared to the traditional approaches with an average improved accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\times \times \%$</tex>. Our study results suggest that RNNs could serve as a reliable algorithm for performance enhancement, potentially improving user experience during BCI operation.