Article Preview
TopSport analytics refers to the use of data and advanced statistics, for example machine learning techniques, to measure performance with the aim to take informed decisions and gain a competitive sports advantage. In other words, sport analytics is the practice of applying mathematical and statistical principles to different sports, such as baseball (Dietrich et al., 2014), basketball (Jain and Kaur,2017) and hockey (Liu and Schulte,2018). In soccer, the most usage is about the prediction of results and the definition of strategies that can be used to win a game or to obtain an improvement of the team performances. Usually, the models constructed in these analysis are based on several aspects about the game, such as tactical, technical or physical information.
However, although each sport has its own characteristics, sport analytics uses the same basic methods and approaches as any other kind of data analysis and, when properly applied, can yield tremendous competitive advantages to a team or an individual player.
The analysis that can be performed with sport analytics is typically divided into two different parts: bio-mechanical and notational analysis (Hughes and Franks,2004). Both techniques involve the analysis and improvement of the sport performance giving good feedbacks to coaches and athletes. Sport biomechanics is concerned with fine detail about individual sport techniques while, on the other hand, notational analysis is more concerned with gross movements or movement patterns in games or teams, and is primarily interested in strategy and tactics. These types of analysis are useful because, if we consider the well- chosen performance indicators to evaluate a specific game, it can be possible to highlight advantageous strategy or important aspect of team performance. In other words, they help coaches to identify good and bad performances of team member or the whole team. (Bartlett,2001).
As we have said before, one of the most widespread use of sport analytics, in soccer environment, is related to the prediction of soccer match results. In literature there are several works focused about the most important factors that influence the results of a game. In (Capobianco et al.,2019), the authors propose a new feature set aimed to model a soccer match. The set is related to characteristics obtainable not only at the end of the match, and it is used to predict the results of the match and the number of goal
scored by the team that won the game. In (Joseph et al.,2006), an approach based on Bayesian Networks to predict match results has been presented. The analysis showed that the Bayesian networks is generally superior to other techniques such as the MC4, a decision tree learner, naive Bayesian learner (NB), and k-nearest neighbor learner (KNN) for this domain in terms of predictive accuracy. Specifically, authors obtain an accuracy equal to 59% which outperformed other machine learning models i.e., 41.7% (obtained by the MC4 classification algorithm), 47.86% (with the NB algorithm) and 50.58% (with the KNN algorithm). A similar analysis has been proposed in (Liti et al.,2017), where the authors predict the outcome of soccer matches finished with a draw at the end of the first half using the information stored during the first part of the match; while, in (Razali et al.,2017) a Bayesian Network approach to predict the outcome of English Premier League matches has been constructed. In (Berrar and Dubitsky,2019) the authors suggest that a key factor in soccer match outcome prediction lies in the successful incorporation of domain knowledge into the machine learning modeling process.
In soccer, there are other types of work concerning specific aspects of the game or player performance analysis. For example, in (Fernandez and Cervone,2019) it has been presented a model that quantifies the expected outcome of a soccer possession at any time during the possession, driven by a fine-grained evaluation of the full spatio-temporal characteristics of the 22 players and the ball. In (Kharrat et al.,2017) the authors try to examine who are the best players in European football, and demonstrate how the players’ ratings evolve over time, using plus-minus rating. Finally, in (Schultze and Wellbrock,2017) it has been proposed a weighted plus/minus metric to be used as an instrument to evaluate player performance.