Artifact Annotations Partially Substitute for Per-User Calibration: SAFE-EDA and a Normalization-Controlled Evaluation of Wrist-EDA Affect Recognition

This is the authors' abstract. We don't add key points for this paper.

SAFE-EDA software for artifact-supervised learning from electrodermal activity. This code-focused release provides model definitions, training and prediction commands, a validated local NPZ loader, seeded PyTorch batching, exact dependency versions, synthetic interface examples, and an architecture illustration.Research data must be obtained directly from EDABE, WESAD, and PhysioNet under their respective terms. This archive contains no participant recordings, prepared research windows, participant identifiers, empirical result tables, per-window evaluation outputs, trained checkpoints, or third-party publication excerpts. Synthetic examples are generated locally and are not experimental evidence.The loader reads prepared normalized windows; raw-data parsing, resampling, fold construction, and normalization must be supplied by the user. Optional statistical and plotting scripts require local metric exports that are not bundled. This archive alone does not reproduce all numerical manuscript results.Associated manuscript: Artifact Annotations Partially Substitute for Per-User Calibration: SAFE-EDA and a Normalization-Controlled Evaluation of Wrist-EDA Affect Recognition. Authors: Haochen Chai, Xinbi Luo, Zining Liu, and Fangfang Jiang.