A multi-season machine learning approach to examine the training load and injury relationship in professional soccer

Aug 5, 2026

This study uses machine learning to quantify how training load relates to soccer injury across multiple seasons in one English Premier League club. Data from 35 male professional players were tracked from the 2014–2015 season through 2018–2019, covering 106 training-load variables across GPS, personal, physical, psychological, and workload models. Models (XGBoost and artificial neural networks) were trained on four and a half seasons and tested on the next half season, with SHAP used for interpretation. Results show high recall for non-contact injury prediction despite class imbalance, identifying key contributing features.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages19

Reading Time48 min

AccessFree to download

Abstract

Objectives

Methods

Results

Conclusions

Keywords

Journal of Sports Analytics 10 (2024) 47–65 DOI 10.3233/JSA-240718 IOS Press 47 A multi-season machine learning approach to examine the training load and injury relationship in professional soccer Aritra Majumdara , 1 ,∗ , Rashid Bakirovb,2 , Dan Hodgesc , Sean McCullaghd and Tim Reesa ,3 a Department of Rehabilitation and Sport Science, Bournemouth University, Fern Barrow, UK b Department of Computing an Informatics, Bournemouth University, Fern Barrow, UK cDan Hodges, Head of Sport Science, AFC Bournemouth, and Head of Performance, Newcastle United FC, UK d Sean McCullagh, First Team Sport Scientist, AFC Bournemouth, UK Received 2 February 2023 Accepted 22 December 2023 Published 22 April 2024 Abstract. OBJECTIVES: The purpose of this study was to use machine learning to examine the relationship between training load and soccer injury with a multi-season dataset from one English Premier League club. METHODS: Participants were 35 male professional soccer players (aged 25.79 ± 3.75 years, range 18–37 years; height 1.80 ± 0.07 m, range 1.63–1.95 m; weight 80.70 ± 6.78 kg, range 66.03–93.70 kg), with data collected from the 2014–2015 season until the 2018–2019 season. A total of 106 training loads variables (40 GPS data, 6 personal information, 14 physical data, 4 psychological data and 14 ACWR, 14 MSWR and 14 EWMA data) were examined in relation to 133 non-contact injuries, with a high imbalance ratio of 0.013 . RESULTS: XGBoost and Artiﬁcial Neural Network were implemented to train the machine learning models using four anda half seasons’ data, with the developed models subsequently tested on the following half season’s data. During the ﬁrst four and a half seasons, there were 341 injuries; during the next half season there were 37 injuries. To interpret and visualize the output of each model and the contribution of each feature (i.e., training load) towards the model, we used the Shapley Additive Explanations (SHAP) approach. Of 37 injuries, XGBoost correctly predicted 26 injuries, with recall and precision of 73% and 10% respectively. Artiﬁcial Neural Network correctly predicted 28 injuries, with recall and precision of 77% and 13% respectively. In the model using Artiﬁcial Neural Network (the relatively more accurate model), last injury area and weight appeared to be the most important features contributing to the prediction of injury. CONCLUSIONS: This was the ﬁrst study of its kind to use Artiﬁcial Neural Network and a multi-season dataset for injury prediction. Our results demonstrate the potential to predict injuries with high recall, thereby identifying most of the injury cases, albeit, due to high class imbalance, precision suffered. This approach to using machine learning provides potentially valuable insights for soccer organizations and practitioners when monitoring load injuries. Keywords: Soccer injury, predictive analytics, machine learning, English premier league, artiﬁcial neural network 1 ORCID: [https://orcid.org/0000-0002-5052-8415](https://orcid.org/0000-0002-5052-8415) . 2 ORCID: [https://orcid.org/0000-0002-2809-9626](https://orcid.org/0000-0002-2809-9626) . 3 ORCID: [https://orcid.org/0000-0001-5498-0145](https://orcid.org/0000-0001-5498-0145) . ∗ Corresponding author: Aritra Majumdar, Department of Rehabilitation and Sport Science, Faculty of Health and Social Sciences, Bournemouth University, Fern Barrow, Poole BH12 5BB, UK. Tel.: +447436937216; E-mail: [amajumdar@bournemouth.ac.uk](amajumdar@bournemouth.ac.uk). 1. Introduction Monitoring the load placed on athletes in training and competition is a current “hot topic”(Kalkhovenet al., 2021) in sport science, with professional sports teams investing substantial resources to this end (Bourdon et al., 2017) . Load monitoring is essen- ISSN 2215-020X © 2024 – The authors. Published by IOS Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial License (CC BY-

FAQs

What is the study’s objective regarding training load and injury?

To examine the relationship between training load and soccer injury using a multi-season dataset and machine learning.

How were the machine learning models trained and evaluated?

XGBoost and artificial neural networks were trained using four and a half seasons of data, then tested on the following half season.

Which approach was used to interpret model outputs and feature contributions?

Shapley Additive Explanations (SHAP) was used to interpret and visualize each model’s predictions and the contribution of training-load features.

Aug 5, 2026

Upload by Ava Thompson