DORA cross-database EEG workload portability: supporting data, code and statistical methods

Description

Supporting materials for “Target-independent verification of electroencephalography workload decoders for adaptive human–machine systems.” This dataset contains analysis code, software environment specifications, a public pre-target analytical specification, complete model-selection traces, non-identifying UNIVERSE prediction scores, participant-level AUC summaries, aggregate statistical results, diagnostics, generated figures, and machine-readable tables for a seven-database EEG workload portability audit. Four controlled-workload databases (RITHM, EEGMAT, COG-PBCI and STEW) supplied development evidence; UNIVERSE supplied untouched external confirmation; MultiPhysio-HRC and SenseCobot supplied joint environment-and-reported-effort stress tests. Third-party raw EEG archives and participant-level derived feature matrices are excluded. Source accessions and licenses are listed in data_sources.tsv. Original package materials are licensed under CC BY 4.0; analysis code is additionally licensed under the MIT License. Version 2 aligns the record and documentation with the current Engineering Applications of Artificial Intelligence submission. Scientific data, analysis code, results, predictions, figures and statistical methods are unchanged from Version 1. The canonical prediction file SHA-256 remains d3be68bf4efa9bd7ef5d72e6bd8604cbbb0dc1ba42cbf935fb34f6dae3df827c.

Steps to reproduce

1. Extract DORA_supporting_data_code_v1.1.zip. 2. From the package root, run `sha256sum -c MANIFEST.sha256` and `python3 verify_frozen_outputs.py` to validate the fixed method, key estimates, prediction hash, evidence roles, and raw-data exclusion. 3. For full reproduction, obtain the seven public source datasets from the accessions in data_sources.tsv, comply with their licenses, and verify the listed checksums. 4. Create the software environment from environment.yml or requirements.txt. 5. Follow RUN_ORDER.md. Complete feature extraction requires approximately 71 GB of working storage. Final outputs are included for immediate numerical verification; raw source signals are not redistributed.

Institutions

• Shanghai University of International Business and Economics

  Shanghai, Shanghai
• Guangzhou University

  Guangdong, Guangzhou

Categories

Funders

• National Social Science Fund of China

  Grant ID: 25BGL287

Licence