Bayesian and Frequentist Motor Imagery EEG Classification: Model Scores, Compute Profiling, and MCMC Convergence Diagnostics

This dataset contains model performance scores, compute profiling metrics, and MCMC convergence diagnostics from a large-scale benchmark study comparing Bayesian and frequentist machine learning pipelines for motor imagery EEG classification. Twenty publicly available MOABB datasets and six pipeline pairs were evaluated, spanning classical feature extraction pipelines (CSP, Tangent Space) and deep learning architectures (ShallowConvNet, DeepConvNet). Performance was assessed across six metrics: AUROC, MCC, NLL, Brier score, ECE, and MCE. MCMC convergence was verified via R-hat, effective sample size, Monte Carlo sampling error, divergences, and visual diagnostics including trace, autocorrelation, and rank plots. Compute profiling data covering training time across all pipelines and datasets is also included.