Analysis of football game-related statistics using multivariate techniques

Abstract

The purpose of this study was to explore football game-related statistics during a competition, using principal component and cluster analyses to determine if it is possible to distinguish the winning teams from the drawing and losing ones. We collected the game-related statistics of the group phase matches of the 2006 World Cup and organised them into a matrix. The principal components of the covariance matrix were calculated. The scores of the first and second components were used to represent the new data, and cluster analysis was applied to separate the elements in two groups (G1 and G2). To analyse the degree of separation between the groups, we calculated the Silhouette Coefficient for each group. Finally, we checked if the winning teams were classified into the same group. The Silhouette Coefficients found for G1 and G2 were 0.54 and 0.55, respectively. Results showed that 70.3% of the winning teams were classified into the same group (G1). Similarly, 67.8% of the drawing and losing teams were classified in G2. This study presented a different way to analyse game-related statistics that allowed the multivariate differences to be shown between successful and unsuccessful teams.

How to be Successful in Football: A Systematic Review

“…Lago et al [48] were the first authors who used a discriminant analysis to identify differences between winners and losers. Moura et al [51] combined this approach with a factor analysis. They investigated 14 variables and performed a factor analysis.…”

Football scout analysis models (based in the 2013/2014 Champions League)

“…Ser capaz de reconhecer essas variáveis permitiria os técnicos entender as ações que podem levar ao sucesso ou ao insucesso em uma partida 6 . Estudos prévios buscaram modelar o jogo de futebol com o objetivo de predizer variáveis menos complexas para explicar o jogo usando diferentes técnicas de redução de dimensionalidade como a análise de componente principal 7 , regressão logística [8][9] , analise discriminante 10-12 e teste de comparações de médias 13 .…”

A Gaussian mixture clustering model for characterizing football players using the EA Sports' FIFA video game system. [Modelo basado en agrupamiento de mixturas Gaussianas para caracterizar futbolistas utilizando el sistema de videojuegos FIFA de EA Sports].

“…For example, Barros, Cunha, Magalhães, and Guimarães (2006) applied PCA to represent and quantify the pitch region used by different football players and, using these analyses, to provide tactical information about the team; and Arruda Moura et al (2013) explored football game-related statistics during football competitions in order to group and distinguish variables related to different game outcomes.…”