Exploring essential variables for successful and unsuccessful football teams in the "Big Five" with multivariate supervised techniques

This research proposes multivariate techniques for discovering the game actions that contribute to the final ranking of football teams. This study uses data from the “Big Five” teams that competed in the Bundesliga First Division, Premier League, LaLiga, Ligue 1, and Serie A in the 2018-2019 season. The principal component analysis is used for outlier detection and for providing an overall preliminary insight. The statistically significant game actions of the top and bottom teams were studied using three supervised multivariate techniques, namely the partial least squares discriminant analysis, random forest and logistic regression. The partial least squares discriminant analysis model best identifies the variables with the most statistically significant contribution to a team's success or failure. The results were compared with those obtained using two-sample univariate tests (such as the Student's t-test or the Mann–Whitney test), demonstrating the advantages of mul tivariate approaches over univariate approaches. The results indicate that the top teams have both offensive and defensive power, and emphasise the high number of attacking actions; in contrast, the bottom teams have weak defences and few offensive actions.