This study uses various datasets to evaluate the performance of feature extraction and classification methods for EEG signals. The EEG signals analyzed in this research are based on taste, odor, and motor imagery, employing novel methods to interpret these complex signals accurately. Three datasets were used in this study: taste-based EEG signals from 10 healthy subjects, odor-based EEG signals from 5 subjects, and motor imagery EEG data from 29 subjects. Feature extraction methods such as Hilbert Transform (HT) for taste, Wavelet Packet Decomposition (WPD) for odor, and HT for motor imagery were applied. Sequential forward and backward search methods were compared with a newly proposed Random Subset Channel Selection (RSCS) method to determine the most effective channels. For the taste dataset, using the RSCS method, an average classification accuracy of 82% was achieved with a significant reduction in the number of channels, demonstrating a 37.9% improvement over using all channels. In the odor dataset, the proposed method achieved an average accuracy of 99.28% for open-nose conditions and 97.49% for closed-nose conditions, with an 86.3% improvement in classification accuracy and an 89.09% reduction in computational complexity. The RSCS method achieved an average accuracy of 81.56% for the motor imagery dataset, showing superior performance compared to sequential methods. The proposed RSCS method outperforms traditional sequential methods by improving classification accuracy and reducing computational complexity across different types of EEG datasets. This method holds promise for enhancing BCI system performance, significantly improving the detection and early diagnosis of neurological conditions.
