Abstract:
|
Understanding how athlete skill changes with age is of immense importance across a range sports, in large part because it can be used to forecast changes future player value. To date, most existing work on aging curves in sports focuses on modeling changes in aggregate measures of player ability, like Wins Above Replacement (WAR). However, different aspects of player skill and ability evolve differently with age. For example, metrics that demand more athleticism and agility tend to degrade more quickly with age, whereas metrics that are less dependent on athleticism (e.g. free throw shooting in basketball) are relatively invariant to aging (or even improve with practice). Different metrics for player ability/skill are also not independent, but rather are a reflection of a combination of different latent player attributes. In this work, we jointly model how multiple metrics evolve with time using a multi-task Gaussian Process. We demonstrate the utility of our method with an analysis of NBA data.
|