Published November 18, 2026 | Version 2.3.0

Walsh-coded and negotiated targets for motor imagery EEG decoding: code and retained results

  • 1. Department of Electrical and Electronics Engineering, Faculty of Engineering, Marmara University, Istanbul, Türkiye
  • 2. Burdur Mehmet Akif Ersoy University, Burdur, Türkiye

Description

Software and retained results for Walsh-coded and negotiated targets for motor imagery EEG decoding: a controlled evaluation of accuracy and confidence, by Salih Çolakoğlu, Mustafa Onat and Nuri Korhan. The accompanying manuscript is submitted to IEEE Access; this software record is not a journal publication.

Version 2.3.0 adds the folder supplementary_data with the three supplementary data files of manuscript version 1.13: participant-level records of Experiments 1–3, the S1/W5 control with record-based summaries, classical-pipeline, resource and equivalence analyses, and three additional sensitivity analyses. Code, retained results and verification commands are unchanged from version 2.2.0 (DOI 10.5281/zenodo.23504655).

The original 1,415 CSP–CNN/EEGNet fits and 800 EEGNet prediction files are retained. Added studies comprise 1,350 Lee2019 EEGNet test records and 54 classical references, a separate 1,350-model 80/20 calibration control, and 1,320 later-day SHU tests from an architecture/output comparison. The completed 270-fit W4 ungated-anchoring control, its independently audited results, and its fresh execution path are included. Its two exploratory accuracy comparisons remain separate from the earlier five-comparison family. Participant groups, information budgets and statistical families remain separate.

Portable prepared-input training tools, scientific source snapshots and no-training prediction-verification commands are provided. Raw EEG recordings, prepared signal caches, model checkpoints and private correspondence are not included. The original code and stored-results archive retain their bytes.

MIT applies to software. The authors' original documentation and generated numerical results use CC BY 4.0. Original BNCI labels retain CC BY-ND 4.0, Lee/OpenBMI labels retain CC0-1.0, and SHU labels retain CC BY 4.0. The primary software-license field does not override this mixed scope. See LICENSING.md, DATA_ATTRIBUTION.md and third-party notices.

The v2.2.0 DOI 10.5281/zenodo.23504655 and the v2.1.0 DOI 10.5281/zenodo.23290274 identify the older archives only. This version is cited independently. The findings do not establish universal superiority, prevention of forgetting, or clinical readiness.

Files

salihcolakoglu/negotiated-targets-mi-eeg-v2.3.0.zip

Files (152.7 MB)

Name Size
md5:190628f138c2062d9a7b254c1af5b207
152.7 MB Preview Download

Additional details

References