A Case-Based Reasoning Approach to Predicting and Explaining Running Related Injuries

When training for endurance activities, such as the marathon, the risk of injury is ever-present, especially for first-time or inexperienced athletes. And because injuries depend on various factors, there is an opportunity to provide athletes with greater levels of support and guidance when it comes to the risks associated with their training. Hence, in this work we propose a case-based reasoning approach to predict injury risk for marathoners and provide actionable explanations so that runners can understand this risk and potentially reduce it. We do this using the type of activity data collected by common training apps, with extended training breaks used as a proxy for injury (in the absence of explicit injury data), and we show how future breaks can be predicted based on the training patterns of similar runners. Furthermore, we demonstrate how counterfactual explanations can be used to highlight those features that are unique to injured runners (those suffering from training breaks) to emphasise training behaviours that may be responsible for higher levels of injury risk for the target runner. We evaluate our work with a dataset of real-world training data by more than 5,000 real marathon runners.