Using Machine Learning to Predict Injury Risk From Athlete Kinetic Patterns

In recent years, the infrastructure and support for young athletes to reach collegiate and professional leagues has rapidly expanded. Such opportunities have increased the emphasis on injury prevention and performance enhancement, prompting collaboration between the sports industry and data analytics. One approach to injury risk analysis is the Sparta Score: a summative quantity introduced by Sparta Science that synthesizes the Load, Explode, and Drive subscores to predict an athlete’s vulnerability to impairment. The research in this paper explores the correlation between Sparta Score and injury risk, utilizing artificial intelligent systems and machine learning to investigate the leading variables in determining one’s score. Random forests were used to calculate feature importance for certain variables, while neural networks were employed in mapping the accuracy of predicting injury risk from combinations of the identified variables. In performing these analyses, it was discovered that there was a relatively weak correlation between Sparta Score and injury risk; rather, a component dissection of the score yielded a higher accuracy in estimating injury risk level. Within this breakdown, Jump Drive proved to be the most influential factor, by a significant margin, highlighting that a complete assessment of elements is more conclusive than a simplified score.