Analysis of changing risk factors and explanation of risk predictions with machine learning for improved hamstring strain injury prevention in Australian football

Professional athletes and organizations can face significant consequences as a result of injury incidents in sports. Therefore, an abundance of studies has been conducted to identify the risk factors in the hope of preventing injuries from occurring in the first place. Hamstring strain injuries (HSIs) are the most frequent injuries in Australian Football League (AFL). Many studies had shown that there are several prominent risk factors for HSIs. However, this finding cannot be identified with any consistency through assessing the risk factors at a single time point, typically the beginning of a season (e.g., in the pre-season) or more frequently throughout the season (e.g., in the pre-season, early in-season and late in-season). Nonetheless, these studies did not consider the potential variability of risk factors across the season. In light of this, it was hypothesised that risk factors may vary depending on the time of the season.

This thesis aims to answer if the risk of hamstring strain injuries in Australian Football can be reduced through a better understanding of the changing risk factors over the course of the season. Despite the study, identifying HSI risk at individual-level remains a challenge. This study aims to explore whether the risk of HSI for individual players can be better understood by explaining the predictions of machine learning (ML) models.

The study utilised recursive feature selection and cross-validation to provide a holistic understanding of important risk factors at different points. Subsequently, counterfactual explanations were effectively generated for players at risk of sustaining HSI.

The study found that non-modifiable risk factors were primarily linked to pre-season injuries, whereas modifiable risk factors were mostly associated with early in-season injuries. Counterfactual explanations and ML models offer a novel perspective in interpreting risk and finding potential solutions.

Overall, this study provides new insights into risk factors associated with HSIs at different time points, as well as offers a solution for interpreting risk at individual-level using ML models and counterfactual explanations. The findings have important implications for researchers and practitioners who seek to mitigate the risk of HSI in the future.