Machine Learning and Wearable Sensors for Fatigue Detection in Sport and Exercise: A Scoping Review of Models, Sensors, Reference Measures and Validation Practices

This record contains the protocol, search strategies and full review dataset of a scoping review that maps which machine learning models, wearable sensors, reference measures of fatigue and model validation methods have been used to detect fatigue in athletes, physically active people and healthy adults performing exercise, and in which sports. The review follows the JBI methodology for scoping reviews and is reported according to PRISMA-ScR. Elicit, PubMed, IEEE Xplore (searched 21 October 2026), Scopus and Web of Science Core Collection (searched 28 October 2026) were searched for open-access studies published from 2010 onwards. Of 2776 records, 78 studies were included. Retrospective registration. The protocol was drafted on 21 October 2026, before screening, and is deposited after screening, data charting and synthesis. All changes made after the first draft are listed with dates in the change log of the protocol file. Files 00_README.md – description of the record and files 01_protocol_and_change_log.pdf – review protocol with the dated change log 02_TableS2_search_strategies.pdf – full search strategies, dates, limits and record counts for all five sources (Supplementary Table S2) 03_review_dataset.xlsx – all screening decisions (AI first pass and human verification), full-text decisions with reasons, charted data for the 78 included studies, synthesis coding and summary counts 04_TableS3_included_studies.xlsx – characteristics of the 78 included studies (Supplementary Table S3) 05_PRISMA_flow_diagram.png – PRISMA-ScR flow diagram Use of AI tools. Elicit (Elicit Research PBC) was used for searching, AI-assisted screening and initial data extraction. Claude (Anthropic) was used for first-pass screening of the Scopus and Web of Science records, data extraction for studies Elicit could not retrieve, a consistency check of the charted data and coding of synthesis categories. All AI decisions and extractions were verified by the reviewers (S.K., P.P.), as documented in the dataset. Abstracts are not redistributed; records are identified by title and DOI.