MFO-SENN: Self-evolving Radial Basis Function Neural Network using Moth Flame Optimization for Motor Imagery task Classification

Classification of EEG brain signals is one of the challenges for researchers because of the non-stationary nature of these signals. In this study, we address two issues: first, the selection of the most informative and discriminative channels, and the second, obtaining the most efficient architecture for the RBFNN model. The selected channels are used to extract the features using Discrete Cosine Transform. A self-evolving radial basis functions neural network is designed that is based on sub-clusters within each motor imagery task class by using the Moth flame optimization (MFO) algorithm. The MFO algorithm produces the number of neurons in the hidden layer, and the center and spread of each neuron. The proposed algorithm is validated using the benchmarked datasets BCI Competition IV 2a and BCI Competition IV 2b. The proposed method outperforms several other methods and successfully classifies motor imagery tasks.