The evolving field of sports analytics is still in the early stages of its adoption. Moreover, soccer analytics utilizing tracking data is even further limited. This research is motivated by Liverpool's integration of a department for throw-in research, assisting them in winning a league title. This research project makes use of the (generously given) German national soccer team (DFB) tracking and event data which includes all player movement during a game, and more specifically, movement before and after a throw-in.
The probability of a throw-in being completed (according to two mutually exclusive definitions) is estimated using various metrics developed using the aforementioned tracking and event data. Binary classification models are used to estimate the completion probability of a given throw-in. The results show that the model provides an encouraging framework of achieving the goal of a universal throw-in metric. Therefore, any given throw-in may be evaluated, providing a meaningful tool to soccer teams, in the footsteps of xG (expected goal) or xPass (expected pass) models.