Advanced feature engineering in Acute:Chronic Workload Ratio (ACWR) calculation for injury forecasting in elite soccer.

Publisher: Public Library of Science Country of Publication: United States NLM ID: 101285081 Publication Model: eCollection Cited Medium: Internet ISSN: 1932-6203 (Electronic) Linking ISSN: 19326203 NLM ISO Abbreviation: PLoS One Subsets: MEDLINE

Original Publication: San Francisco, CA : Public Library of Science

Soccer*/injuries ; Workload* ; Athletic Injuries*/prevention & control ; Athletic Injuries*/epidemiology; Humans ; Forecasting ; Machine Learning

Competing Interests: The authors have declared that no competing interests exist.
Controlling training monotony and monitoring external workload using the Acute:Chronic Workload Ratio (ACWR) is a common practice among elite soccer teams to prevent non-contact injuries. However, recent research has questioned whether ACWR offers sufficient predictive power for injury prevention in elite competition settings. In this paper, we propose a novel feature engineering framework for training load management, inspired by bilinear modeling and signal processing principles. Our method represents external workload variables, derived from GPS data, as discrete time series, which are then integrated into a temporal matrix termed the Footballer Workload Footprint (FWF). We introduce calculus-based techniques-applying integral and differential operations-to derive two representations from the FWF matrix: a cumulative workload matrix ([Formula: see text]) generalizing Acute Workload (AW), and a temporal variation matrix ([Formula: see text]) generalizing Chronic Workload (CW) and formulating the ACWR. Our approach makes traditional workload metrics suitable for modern machine learning. Using real-world data from an elite soccer team competing in LaLiga (Spain's top division) and UEFA tournaments, we conducted exploratory and confirmatory analyses comparing multivariate models trained on FWF-derived features against those using traditional ACWR calculations. The FWF-based models consistently outperformed baseline methods across key performance metrics-including the Area Under the ROC Curve (ROC-AUC), Precision-Recall AUC (PR-AUC), Geometric Mean (G-Mean), and Accuracy-while reducing Type I and Type II errors. Tested on temporally independent holdout data, our top model performed robustly across all metrics with 95% confidence intervals. Permutation tests revealed a significant association between FWF matrices and injury risk, supporting the empirical validity of our approach. Additionally, we introduce an interpretability framework based on heatmap visualizations of the FWF's cumulative and temporal variations, enhancing explainability. These findings indicate that our approach offers a robust, interpretable, and generalizable framework for sports science and medical professionals involved in injury prevention and training load monitoring.
(Copyright: © 2025 Matas-Bustos et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.)

Int J Sports Physiol Perform. 2019 Apr 1;14(4):544-546. (PMID: 30702360)
J Sci Med Sport. 2017 Dec;20(12):1068-1074. (PMID: 28595869)
Philos Trans A Math Phys Eng Sci. 2021 Oct 4;379(2207):20200363. (PMID: 34398656)
BMJ Open Sport Exerc Med. 2019 Oct 30;5(1):e000573. (PMID: 31798948)
Imeta. 2022 Aug 01;1(3):e43. (PMID: 38868715)
Br J Sports Med. 2019 Aug;53(16):988-990. (PMID: 29807930)
Sports Med Open. 2022 Jun 7;8(1):73. (PMID: 35670925)
PLoS One. 2019 Feb 14;14(2):e0211776. (PMID: 30763328)
PLoS One. 2015 Mar 04;10(3):e0118432. (PMID: 25738806)
Sci Rep. 2024 Mar 13;14(1):6086. (PMID: 38480847)
PLoS One. 2018 Jul 25;13(7):e0201264. (PMID: 30044858)
Br J Sports Med. 2016 Jun;50(12):725-30. (PMID: 26795611)
Br J Sports Med. 2021 Oct;55(20):1170-1178. (PMID: 34001503)
Br J Sports Med. 2020 Apr;54(7):421-426. (PMID: 31182429)
Br J Sports Med. 2022 Dec 6;:. (PMID: 36588400)
J Sci Med Sport. 2017 Jun;20(6):561-565. (PMID: 27856198)
J Sports Sci. 2012;30(2):121-7. (PMID: 22122431)
Br J Sports Med. 2016 Mar;50(5):273-80. (PMID: 26758673)
Cancer Genomics Proteomics. 2018 Jan-Feb;15(1):41-51. (PMID: 29275361)
Nature. 2020 Sep;585(7825):357-362. (PMID: 32939066)
J Athl Train. 2016 May;51(5):410-24. (PMID: 27244125)
IEEE Trans Syst Man Cybern B Cybern. 2009 Apr;39(2):539-50. (PMID: 19095540)
Sports Med. 2021 Mar;51(3):581-592. (PMID: 33332011)
Nat Methods. 2020 Mar;17(3):261-272. (PMID: 32015543)
Med Sci Sports Exerc. 2018 Nov;50(11):2267-2276. (PMID: 29933352)
Sports Med Open. 2022 Jan 25;8(1):15. (PMID: 35076796)
J Exp Orthop. 2021 Apr 14;8(1):27. (PMID: 33855647)
Sci Med Footb. 2023 Aug;7(3):214-228. (PMID: 35757889)
Int Conf Affect Comput Intell Interact Workshops. 2013;2013:245-251. (PMID: 25574450)
Br J Sports Med. 2017 May;51(9):749-754. (PMID: 28003238)
Br J Sports Med. 2016 Apr;50(8):444-5. (PMID: 26795610)
Br J Sports Med. 2017 Feb;51(3):209-210. (PMID: 27650255)
JAMA. 2013 Nov 27;310(20):2191-4. (PMID: 24141714)
Sports Med. 2016 Jun;46(6):861-83. (PMID: 26822969)
J Sci Med Sport. 2013 Nov;16(6):556-61. (PMID: 23333009)
Sports Med. 2020 Mar;50(3):561-580. (PMID: 31691167)
Front Physiol. 2020 Aug 20;11:944. (PMID: 32973542)
Br J Sports Med. 2016 Feb;50(4):231-6. (PMID: 26511006)
Br J Sports Med. 2020 Jun;54(12):731-738. (PMID: 30792258)
IEEE Trans Neural Netw. 2000;11(5):1188-93. (PMID: 18249845)
Clin Exp Pharmacol Physiol. 2002 Mar;29(3):218-22. (PMID: 11906487)
Med Sci Sports Exerc. 1998 Jul;30(7):1164-8. (PMID: 9662690)
Int J Sports Physiol Perform. 2015 May;10(4):489-97. (PMID: 25393111)
Int J Sports Physiol Perform. 2017 Apr;12(Suppl 2):S22-S28. (PMID: 28253038)
PLoS One. 2022 Jul 14;17(7):e0270099. (PMID: 35834441)

Date Created: 20250723 Date Completed: 20250723 Latest Revision: 20250731

20250731

PMC12286412

10.1371/journal.pone.0327960

40700417