Abstract
Brain Computer Interfaces (BCIs) will convert the thoughts of individuals with physical disabilities into commands for devices to enable them autonomous mobility. The Electroencephalogram (EEG) is widely favoured as a control signal due to its ease of acquisition compared to invasive recordings. While the affordability of EEG equipment allows for the use of numerous recording channels, this abundance increases computational complexity, necessitating optimal channel selection strategies to improve efficiency and classification accuracy. Deep Neural Networks (DNNs) often face scalability issues with multidimensional, locally correlated inputs, making them impractical for such applications. Convolutional Neural Networks (CNNs) are efficient for analysing BCI data but require careful hyperparameter tuning to achieve optimal performance. This paper introduces a framework for classifying BCI channel selection using deep learning techniques. The study primarily concentrates on refining the hyper parameters of deep learning algorithms through metaheuristic techniques, specifically employing Discriminated Stochastic Diffusion Search (SDS) to enhance BCI channel selection. The findings indicate that the proposed hyperparameter optimization methods, such as Discriminated-SDS, significantly enhance classification accuracy. The proposed D-SDS balances exploration and exploitation, mitigates the local optima issue, and is especially advantageous for intricate deep learner architectures such as VGGNet, ResNet, and InceptionNet. Hyperparameter optimization in EEG-based BCI systems can substantially improve performance, enhancing their efficiency and reliability.
Keywords: BCI Channel Selection, Deep Learning, Discriminated-SDS, Electroencephalogram (EEG), HyperParameter, Stochastic Diffusion Search (SDS).