Joint MVMD-based optimal feature selection and FW-LS-TWSVM for motor imagery recognition.

The Motor Imagery-Brain Computer Interface (MI-BCI) system is an effective approach for motor neurorehabilitation training and human-machine collaborative control. However, the current MI-BCI systems' decoding accuracy and real-time performance still fall short of practical requirements. To address this issue, this study proposes a model combining MVMD-based optimal feature selection and the Fuzzy Weighted Least Squares Twin Support Vector Machine (FW-LS-TWSVM). First, raw data is decomposed into multiple Intrinsic Mode Functions (IMFs) using Multivariate Variational Mode Decomposition (MVMD). Then, Common Spatial Pattern (CSP) is employed to extract features from each IMF, and a feature selection method based on F-statistics is used to adaptively identify the optimal IMFs and their corresponding features, thereby extracting optimal frequency information. Subsequently, this paper introduces, for the first time, the application of the FW-LS-TWSVM to MI-BCI EEG decoding, aiming to enhance the identification efficiency of outliers. The proposed method was validated on two publicly available motor imagery datasets, achieving accuracies of 87.40% and 88.48%, respectively. Comparative analysis revealed that both the frequency band decomposition method and the FW-LS-TWSVM classification model contributed significantly to the decoding accuracy. Compared to traditional frequency band decomposition, SVM, and its improved variants, the proposed method not only achieved higher accuracy but also required relatively less training time. These results indicate that the proposed model can facilitate the development of MI-BCI systems, enhance the behavioral capabilities of healthy individuals, and help improve the quality of life for patients with neurological disabilities.