Overview

• The interpretable model, led by Daniel Acuña and published Aug. 27 in Science Advances, was trained on DOAJ data including 12,869 vetted journals and 2,536 titles removed for quality violations.
• When run on 15,191 open-access journals in Unpaywall, it flagged 1,437 titles, with an estimated 345 false positives, leaving more than 1,000 judged questionable after expert checks.
• Researchers also estimate the tool failed to catch about 1,782 problematic titles, underscoring that it is not a final arbiter of journal quality.
• Signals the system weighs include short publication turnaround, high self-citation, unclear editorial policies, missing or weak editorial boards, website quality issues, and licensing or fee opacity.
• The study does not name specific journals due to legal concerns, notes that many flagged titles were not on watchlists and include some from large publishers, and says the tool is not publicly available, with limited testing reported for indexers and publishers.