A Frequency-Shifting Variational Mode Decomposition-Based Approach to MI-EEG Signal Classification for BCIs

Electroencephalogram (EEG) signal analysis is crucial for understanding neural activity and advancing diagnostics in neurology. However, traditional signal decomposition (SD) techniques are hindered by two critical issues, mode mixing and mode aliasing, that compromise the quality of the decomposed signal. These challenges result in poor signal integrity, which significantly affects the accuracy of subsequent EEG interpretations and classifications. As EEG analysis is widely used in diagnosing conditions such as epilepsy, brain injuries, and sleep disorders, the impact of these shortcomings can be far-reaching, leading to misdiagnoses or delayed treatments. Despite extensive research on SD techniques, these issues remain largely unresolved, emphasizing the urgent need for a more reliable and precise approach. This study proposes a novel solution through the frequency-shifting variational mode decomposition (FS-VMD) method, which overcomes the limitations of traditional SD techniques by providing better resolution of intrinsic mode functions (IMFs). The FS-VMD method works by extracting and shifting the fundamental frequency of the EEG signal to a lower frequency range, followed by an iterative decomposition process that enhances signal clarity and reduces mode aliasing. By integrating advanced feature selection techniques and classifiers such as support vector machines (SVM), convolutional neural networks (CNN), and feature-weighted k-nearest neighbors (FWKNN), this approach offers a significant improvement in classification accuracy, with SVM achieving up to 99.99% accuracy in the 18-channel EEG setup with a standard deviation of 0.25. The results demonstrate that FS-VMD can address the critical issues of mode mixing and aliasing, providing a more accurate and efficient solution for EEG signal analysis and diagnostics.