Injury risk and performance: Towards a better understanding of the complexities and intricacies of load monitoring within an elite football club.

Abstract

Load monitoring has emerged as a pivotal aspect of contemporary sports science, particularly in the context of athlete training and competition. This thesis delves into the dynamic landscape of load monitoring with a particular focus on soccer, a sport of unparalleled global popularity, boasting 200,000 professional and 240 million amateur players. The prevalence of soccer- related injuries, surpassing those in other sports, has underscored the imperative for effective load monitoring strategies to optimize training adaptations, evaluate fatigue and recovery, and mitigate injury risks. Professional sports teams, cognizant of the multifaceted implications of inadequate load management, have invested significantly in this domain. In the realm of soccer, where injuries can lead to prolonged player absences, impacting team performance and incurring substantial financial costs, the need for comprehensive load monitoring becomes even more apparent. Notably, English Premier League soccer clubs bore an approximate financial burden of £45 million per season due to injuries from 2012-2013 through to the 2016- 2017 season. In response to the pressing demand for a nuanced understanding of the intricate relationship between training load and soccer injuries, this thesis integrates insights from machine learning. Building upon existing research, we explore how machine learning techniques contribute to the refinement of load monitoring strategies in soccer, offering a promising avenue for enhancing injury prevention protocols. By bridging the gap between traditional sports science methodologies and cutting-edge machine learning applications, this research seeks to provide a comprehensive framework for optimizing athlete performance and well-being in the dynamic context of soccer with the help of Machine learning. This thesis undertook three comprehensive investigations aimed at advancing the understanding of the relationship between training load and soccer injuries through the application of machine learning methodologies. The initial inquiry critically examined recent research endeavours in football that incorporated machine learning techniques. This exploration highlighted the profound implications of football injuries, which not only result in prolonged player absences affecting team performance but also entail considerable financial ramifications. Despite the burgeoning interest in the relationship between training load and injuries, prevailing models and statistical approaches were found to inadequately capture the intricate nuances of this association. The lack of consensus on variables for analysis posed a significant challenge, hindering the effective utilization of existing studies in guiding the selection of key training load variables. (Chapter – 2)1 Subsequently, the second investigation employed machine learning to scrutinize the connection between training load and soccer injuries, utilizing a multi-season dataset from an English Premier League club. A pioneering aspect of this chapter was the application of Artificial Neural Networks, marking the first instance of employing such a method on a multi-season dataset for injury prediction. The results indicated a promising capability to predict injuries with high recall, identifying a majority of injury cases. However, precision suffered due to the prevalent class imbalance, emphasizing the need for further refinement in this methodology. Despite these challenges, the chapter provided valuable insights for soccer organizations and practitioners engaged in load injury monitoring. (Chapter – 3)2 The third and final investigation contributed a pioneering analysis of online continual and adaptive learning methodologies for soccer injury prediction, utilizing a distinctive multi- season dataset from Elite Premier League players. Noteworthy findings demonstrated the superiority of these adaptive learning approaches over static learning, with cumulative training identified as a critical factor enhancing model adaptability and performance. The practical applications extended to injury prevention and player well-being management in professional soccer. The research's forward-looking stance emphasized the necessity for future exploration into advanced continual learning frameworks and real-time injury prediction systems to refine and enhance the efficacy of injury prevention strategies. (Chapter – 4)3 ______ 1 The second chapter is published as a journal paper entitled “Machine Learning for Understanding and Predicting Injuries in Football” in the Sports Medicine – Open Journal (SMOA). 2 The third chapter is published as a journal paper entitled “A Multi-Season Machine Learning Approach to Examine the Training Load and Injury relationship in Professional Soccer” in the Journal of Sports Analytics (JSA). 3 The fourth chapter, entitled as “A Multi-Season Continual Machine Learning Approach to Examine the Training Load and Injury relationship in Professional Soccer” is ready for submission.