A simple but tough-to-beat baseline for fMRI time-series classification

Pavel Popov, Usman Mahmood, Zening Fu, Carl Yang, Vince D. Calhoun, Sergey M. Plis

Abstract: Highlights•A surprisingly simple MLP model outperforms complex AI models in fMRI classification.•This simple MLP generalizes to unrelated datasets posing the same prediction task.•These findings hold across many datasets, disorders, and processing pipelines.•fMRI dynamics may contain much less discriminative information than commonly believed.•Predictive accuracy alone may not justify using complex AI models with fMRI data.