Improving the classification performance of eeg signals with a novel random subset channel selection approach: Applications on taste, olfactory, and motor imagery datasets

Improving the performance of Brain-Computer Interface (BCI) systems and enabling the early diagnosis of neurological diseases have become critical research topics in contemporary neuroscience and neural engineering. This study uses various datasets to evaluate the performance of feature extraction and classification methods for EEG signals. Three different datasets were examined: taste-based EEG signals from 10 subjects, odor-based EEG signals from 5 subjects, and motor imagery EEG data from 29 subjects. Hilbert Transform (HT) was applied for taste data, Wavelet Packet Decomposition (WPD) for odor data, and HT for motor imagery data. The proposed Random Subset Channel Selection (RSCS) method was compared with sequential search methods, and the most effective channels were identified. The RSCS method achieved 82% accuracy for the taste dataset while reducing computational complexity by 37.9%. For the odor dataset, 98.38% accuracy was achieved, with an 89.09% reduction in computational complexity. RSCS achieved 81.56% accuracy for the motor imagery dataset, outperforming sequential methods and reducing computational complexity by 75%. In conclusion, the proposed RSCS method enhances classification accuracy while reducing computational complexity compared to traditional sequential methods. This method holds the potential for improving the performance of BCI systems, particularly for EEG signals, and offers significant contributions to the detection and early diagnosis of neurological disorders.