Clustering soccer players to find the drivers of soccer team performance

In the professional soccer business, technical directors are under constant pressure to perform well with their teams. The choices made in assembling a team lineup are of crucial importance to achieve satisfying results. This paper aims to answer three interconnected research questions to investigate how team lineups in soccer matches should be formed to achieve positive match outcomes. By first assigning soccer players to different player types based on the outcomes of several cluster algorithms, the attributes distinguishing one soccer player from another are identified. Subsequently, the presence of combinations of the resulting player types are found with the use of rare correlated pattern mining. Finally, by estimating match outcomes with an ordered probit model, the effect of the presence of a certain combination of player types in the lineup of a team is investigated. The results of the match outcome estimation show that a good goalkeeper is particularly important for a soccer team in away matches. Another interesting finding is that although player types on their own can negatively influence the match outcome, a combination of them may positively influence the match outcome.