How can success in soccer be measured? With the amount of positional data available in soccer today, this question seems to be particularly interesting, especially prior to a world championship. Private Lecturer Dr. Daniel Link from the Chair for Training Science and Sport Informatics (Prof. Dr. Martin Lames) developed a model with which a team's danger of scoring a goal can be measured.
Goals in soccer alone can only to a limited degree be used to make a statement about the performance of a team and the quality of its players: They occur only infrequently, and occasionally due to a single moment of inattentiveness or when a highly-dominant team lacks a bit of luck. Such indicators as shots on goal, successful passes, individual duels, ball possession and distances covered are commonly cited, particularly in the media, although their benefit as a unit for performance must be questioned. In the semi-final of the last world championship of 2014, for example, Germany had fewer goals than Brazil (in a relationship of 14 to 18), but hardly a spectator would express doubts on the superiority of Germany in this game (7 to 1 goals).
In his post-doctoral dissertation on "Data Analytics in Professional Soccer", Link summarized various journal articles in which the performance of soccer teams is examined. A central criterion is the situations in which the possibility exists for making a goal. "In soccer, this depends primarily on the fact that a team with the ball comes into a range where there is the danger of making a goal or where the opponent can prevent this from occurring," says the scientist from the Chair for Training Science and Sport Informatics. In the just published work, he presents six individual studies with innovative mathematical approaches for play analysis and player evaluation in professional soccer.
Real-time analysis with the aid of optical tracking
In chapter 3, which was originally published in "PLOS One", the training scientist presents objective criteria for determining the team's performance in real-time using an algorithm designed especially for determining this. With this procedure, he determines a quantitative representation of the probability that a goal can be made by a player in possession of the ball at any point in time - "I call this dangerousity," says Link. The calculation of these metrics is based on the spatial constellation of a player and a ball, and is made up of the four components of pressure, density, ball control and playing field zone. The English term dangerousity has, meanwhile, become referenced world-wide in the sport data community.
With this approach, the author in co-operation with the German soccer league (DFL) evaluated 64 Bundesliga games. The positional data of the players and the ball were recorded using an optical tracking system. In addition, the automatic evaluation of hundreds of game scenarios using Link's algorithm was compared with evaluations made by semi professional soccer coaches. Hereby, a high degree of agreement in the estimation of the individual scenes was to be found between the machine and humans. The evaluation took place with support of the DFL subsidiary company Sportec Solutions (STS).
"From the danger, further metrics can also be derived with which questions can be solved for the analysis of the games," explains the sport scientist regarding his new approach. "We use these metrics in order to analyze individual actions in a game, to describe passages in the game and to characterize the performance and efficiency of teams over the course of a season." For future studies, they would employ a random and a result-independent criterion in order to examine the influence of central incidents, different playing systems or tactical group concepts in a soccer game on success.
Contact:
PD Dr. Daniel Link
Technical University of Munich
Chair for Training Science and Sport Informatics
Tel: +49.89.289.24498
E-Mail: daniel.link(at)tum.de
Text: Sabine Letz