This study evaluates whether any single decoding pipeline dominates across subjects in motor imagery brain-computer interfaces by testing 1,056 configurations on three public datasets using rigorous statistical benchmarks.
- Evaluated >340,000 subject-level model fits across PhysionetMI, Cho2017, and Zhou2016 datasets within the MOABB framework.
- Applied Friedman omnibus tests, Nemenyi critical-difference analysis, and Wilcoxon signed-rank tests to compare feature extractors, scalers, and classifiers.
- Covariance tangent-space projection (cov-tgsp) and Common Spatial Patterns (CSP) were the strongest families but showed dataset-dependent ordering.
- On the PhysionetMI cohort, the best pipelines were statistically indistinguishable (Nemenyi p = 0.27; Kendall's W = 0.11).
- The single best pipeline was optimal for only 35% of participants, while nonlinear descriptors were best for roughly one third.
- Matching the pipeline to the participant improved accuracy by approximately seven points over the best fixed choice.
The findings indicate that no universal decoder exists even under favorable conditions, providing a quantitative case for participant-aware model selection rather than relying on average rankings.