Markov Decision Processes with Contextual Nodes as a Method of Assessing Attacking Player Performance in Rugby League

Although numerous statistics have been generated across several sports to evaluate player performances, they are generally limited by failure to understand different contexts, or the impact of actions on future point scoring opportunities. Despite recent advances in ice hockey incorporating both context and lookahead into the analysis of valuable actions for individual players, so far there is no such research within rugby league. In this study, we aim to address this problem using a dataset from the 2020 Super League season. We use a Markov Decision Process framework to evaluate the Q-value of attacking actions within rugby league. The Q-value provides the value of each state-action pair as they pertain to the goal of scoring points in rugby league. We compare the Q-value to the state value to define action impact ratings which are used to evaluate players’ performances. Our results show that context, in the form of the period of the match or current match score, can provide additional information with regards to a player’s performance ratings.