Investigating Optimal EEG Channels and Features for Brain-Computer Interfaces: An Exploration using Evolutionary Algorithms and Machine Learning

Optimal EEG Channels and Features in BCI

Abstract

Brain-Computer Interfaces (BCI) are advanced systems that enable a direct neural pathway between the human brain and external devices. The importance of BCI is underscored by its profound implications for medical therapeutics, particularly in neurorehabilitation. This study developed an algorithm to detect 8 motion commands for a robot using individuals' EEG signals (Electroencephalogram). These signals were recorded during imagined and expressed commands. The research aimed to identify optimal features for extracting and classifying EEG signals for robot commands and to pinpoint the best EEG channels for a cost-effective, efficient signal acquisition system. Four categories of features, including temporal, frequency, wavelet, and combined features, were extracted from the EEG signals. The Imperialist Competitive Algorithm (ICA) and Cuckoo Optimization Algorithm (COA), were utilized for feature selection. Findings revealed that wavelet features are most effective for analyzing and classifying EEGs. For imagined commands, optimal features from all channels achieved a 96.3% classification accuracy, while expressed commands reached 96.5%. The frontal and parietal lobes were identified as the prime EEG channels for command detection, achieving accuracies of 91.5% and 86.9% for imagined commands, and 92.7% and 86.1% for expressed commands, respectively. The result also indicated that the brain's midline and left hemisphere (containing the Broca area) outperformed the right hemisphere in classification. By focusing on the optimal EEG channels, a more cost-effective hardware system can be designed, surpassing the traditional 21-channel system and requiring only 14 electrodes in the frontal and parietal regions.