Automatic EEG channel selection for multiclass brain-computer interface classification using multiobjective improved firefly algorithm

Multichannel Electroencephalography-based Brain-Computer Interface (BCI) systems facilitate a communicating medium between the human brain and the outside world. BCI systems aim to translate human intentions into computer-based control commands by decoding the respective brain patterns. Moreover, Electroencephalography (EEG) analysis involves high-dimensional features that increase the computational burden of applied signal processing approaches. To minimize this overload caused by a large set of channels, we propose an automatic EEG channel selection method for multiclass Motor-Imagery (MI) classification. In this study, we developed a hybrid channel ranking procedure using Fisher information and the objective Firefly Algorithm (FA). Firstly, the preprocessed neural signals are used to extract spatial-temporal features using the Regularized Common Spatial Pattern with Aggregation (RCSPA) method. Then, objective FA with two input variables (Spectral Entropy and Lyapunov exponent) is used to compute a weighted score for each channel in the neighborhood of a candidate solution. Finally, a novel Channel Set Relevance Index (CSRI) is developed to rank channels using their respective weighted score and Fisher information. The RCSPA features of highly ranked channels are employed to discriminate different MI-tasks using the Regularized Support Vector Machine (RSVM) classifier. The proposed approach is cross-validated on three publicly available BCI competition datasets (BCI Competition IV- 2008 - IIA, BCI Competition IV- dataset 1, BCI competition III - dataset IVa) with varying numbers of channels. The validation results show that the proposed method achieved a superior classification accuracy (83.97% on dataset 1, 80.85% on dataset 2, and 84.19% on dataset 3) with fewer channels than other baseline methods. In addition, our method significantly reduced the BCI preparation time, making it effective to conduct multiple experimental sessions for a large pool of subjects.