COMPUTATIONAL MODELING OF PASS EFFECTIVENESS IN SOCCER

Abstract

The emerging data explosion in sports field has created new opportunities to practice data science and analytics for deeper and larger scale analysis of games. With collaborating and competing 22 players on the field, soccer is often considered as a complex system. More specifically, each game is usually modeled as a network with players as nodes and passes between them as the edges. The number of passes usually define the weight of each edge, and these weights are employed to identify the key players using network modeling theory. However, the number of passes metric considers each pass the same and cannot differentiate players who are making ordinary passes, usually in their own pitch to a close teammate, from those who make key passes that start or improve an attack. As a solution, in this paper, we present a descriptive model to quantify the effectiveness of passes in soccer to differentiate between key passes and regular passes with not much contribution to the play of a team. Our model captures the perception of domain experts with a careful combination of risk and gain assessments. We have implemented our model in a soccer data analytics software. We performed a user study with domain experts, and the results show that our model captures domain expert evaluations of a number of example scenarios with 94% accuracy. The proposed model is not computationally demanding which allows real-time pass assessment during games on commodity hardware as demonstrated by our software prototype.