Frequency band choice and the incremental value of joint EEG representations for cognitive task classification
Mean accuracy across frequency bands can obscure whether a comparison method reaches similar performance in its best band. We examined existing MFCSync results for electroencephalographic classification of spatial cognition and episodic memory in 25 and 22 healthy young adults. Seven bands were evaluated separately using within-subject, trial-grouped five-fold validation. Gains over the stronger component, calculated from the reported means, were 3.50 and 2.47 percentage points. Comparing the lowest fusion accuracy with the highest accuracy among 49 nonfused configurations per task left margins of 1.40 and 0.50 points. We decomposed this reduction into the increase in the comparison score and the decrease in the fusion score relative to their reported means. Positive margins also preserved the ordering under all separately chosen convex summaries of the reported accuracies. Strict separation is guaranteed if each accuracy entry changes by an absolute amount below 0.70 and 0.25 points, respectively. Within the evaluated configurations, favorable band selection did not eliminate the fusion advantage, although the episodic-memory margin was small and its sampling uncertainty remains unresolved.
Authors
- Dan Shao (ORCID: https://orcid.org/0000-0003-0493-0769)
- Caixia Meng (ORCID: https://orcid.org/0009-0009-8371-2735)
- Peng Zhang
Institutions
- Zhengzhou Police University (CN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-21
- DOI
- https://doi.org/10.21203/rs.3.rs-11584956/v1
- Primary Topic
- EEG and Brain-Computer Interfaces
- Type
- preprint