Assessing the effectiveness of spatial PCA on SVM-based decoding of EEG data

This study evaluates whether spatial principal component analysis (PCA) improves decoding accuracy in multivariate EEG analysis using support vector machines (SVMs). Several PCA strategies were compared—group-based vs. subject-based decomposition, with and without Varimax rotation—across nine EEG datasets, including ERP components like P3b, N400, and N2pc. The findings reveal that none of the PCA methods consistently improved decoding performance and often reduced accuracy compared to raw channel data. These results caution against the uncritical use of PCA in EEG decoding pipelines.

Recommended citation: Zhang, G., Carrasco, C. D., Winsler, K., Bahle, B., Cong, F., & Luck, S. J. (2024). "Assessing the effectiveness of spatial PCA on SVM-based decoding of EEG data." NeuroImage, 293, 120625. https://doi.org/10.1016/j.neuroimage.2024.120625
Download Paper