Player Roles: How to find the right type of player for your team?

The Player Roles module rates each player in terms of their fit with the 22 different roles that are common in modern football. Based on the terminology used by football experts, the mainstream football media and football video games, we define each player role as a set of characterizing tasks and duties during matches. For instance, we define a Target Man as a striker who excels in aerial duels, holds on the ball higher up the pitch to bring additional teammates into play and converts crosses with powerful headers. The figure below shows the 22 different roles for six different positions, where roles for the same position are in the same color and arrows connect roles that have tasks and duties in common.

Figure 1: Overview of the 22 player roles, where roles for the same position are in the same color and arrows connect roles that have tasks and duties in common. Goalkeeper roles are in orange, central defender roles are in green, wing defender roles are in red, midfielder roles are in dark blue, winger roles are in light blue and striker roles are in purple

To provide more insight into what constitutes a player role, we discuss the characterizing tasks and duties for the Inside Forward, Deep-Lying Playmaker and Ball-Playing Keeper roles in further detail using three illustrative examples.

1. An Inside Forward, such as PSV winger Hirving Lozano, has good ball control (i.e., technique), often dribbles (preferably inside) with the ball taking on his direct opponent, and regularly attempts long-range shots. He occasionally attempts crosses, high-risk passes, and fast combination passes with teammates. He is able to utilize the space behind the defense and is able to find pockets of space both in the danger zone (i.e., the zone in front of the opposing box) and the opposing box, where he attempts to score from close range.
2. A Deep-Lying Playmaker, such as Chelsea midfielder Jorginho, has good ball control and passing technique, dictates the tempo of the game, and prefers to pass the ball to teammates who are higher up the pitch, regardless of the distance between them. He will occasionally attempt to take on opposing players, attempt high-risk passes, and is sometimes involved in fast combinations. In defense, he keeps his position to reduce the space available for the opposing attacking midfielder.
3. A Ball-Playing Keeper, such as Bayern Munich goalkeeper Manuel Neuer, has good ball control, distributes the ball to his teammates both nearby and further up the pitch, and often serves as an eleventh outfield player to deal with teams putting pressure on the defensive line.

Methodology

We rate a player in terms of each of the 22 player roles by automatically analyzing the data that is collected during matches. In order to provide representative player role ratings for a player at each point in time, we restrict our analysis to the matches that the player played during the past eight months. In addition, we only display player role ratings for players who played at least 700 minutes during this period. These restrictions allow us to display reliable player role ratings for over 20,000 different players in the SciSports platform.

To obtain the player role ratings, we analyze play-by-play match event data that describe all on-the-ball actions that occur during matches such as passes, shots, dribbles, crosses, interceptions and tackles. For each action, we consider relevant properties such as the start and end location, the time in the match, the player who performed the action, and type-specific characteristics (e.g., whether a free-kick was taken directly or indirectly).

We developed a machine learning model to determine whether a particular player fits a given player role. We trained our machine learning model on the performances of players in matches played during the 2017/2018 season. To this end, we collected a large number of illustrative examples for each of the 22 player roles (i.e., sets of players who excel in a certain player role) and constructed a large set of statistics that measure how well a player performs the characterizing tasks and duties for each role.

Collecting illustrative examples

We obtained a set of 1,235 illustrative examples for the 22 player roles from a panel of football experts. We presented our panelists with players from the English Premier League, Spanish LaLiga, German 1. Bundesliga, Italian Serie A, French Ligue 1, Dutch Eredivisie and Belgian Pro League. We invited them to assign one of the 22 player roles to each player they had watched often enough during the 2017/2018 season and felt comfortable enough to judge. We used the labeled players to train and evaluate our machine learning model.

Constructing characterizing statistics

We extracted a set of 515 advanced statistics that capture how well a player performs the characterizing tasks and duties for each of the 22 player roles. These statistics include, among others, the number of saves from short-range goal attempts, the number of goalkeeper interceptions outside the penalty area, and the number of passes received in specific areas of the pitch. To obtain a set of reliable advanced statistics, we had to address the following four challenges.

1. Descriptive statistics versus discriminative statistics.
   The statistics do not only need to be able to describe the characteristics of the different player roles, but they also need to be able to distinguish between them. For instance, although the Mobile Striker (e.g., Lionel Messi) and Target Man (e.g., Mario Gómez) roles are quite different, they still have several tasks and duties in common.
2. Quantity of actions versus quality of actions.
   The statistics need to distinguish between the quality and quantity of the actions that a player performs. For instance, the quality of a cross, which is considered successful if it reaches a teammate, does not only depend on the crosser, but also on the receiver. A perfectly-placed high cross is more likely to be converted by the tall Peter Crouch than by the much shorter Lionel Messi.
3. Differences between players within the same team.
   The statistics need to account for differences between players within the same team. For instance, a player who takes two shots from outside the penalty area out of every 25 actions is likely to have a different player role than a teammate who takes two similar shots out of every 150 actions.
4. Differences between similar players in different teams.
   The statistics need to account for differences between similar players in different teams. For instance, Chelsea Deep-Lying Playmaker Jorginho performs 83.5 passes per 90 minutes, whereas Newcastle United Deep-Lying Playmaker Jonjo Shelvey performs 47.9 passes per 90 minutes only.

Use case

The SciSports platform allows to scout interesting players based on data collected during matches. The available filters for, among others, SciSkill, Potential, market value, age and position enable to reduce the list of players to a shortlist in just a blink of an eye. The newly released Player Roles module takes the filter capabilities of the platform one step further. The module provides insights into which roles a player has fulfilled during the past eight months and also enables to find players who are similar to a reference player in terms of player roles. This additional filter reduces the number of shortlisted players even further and enables scouts to spend more time on video analysis and watching players live in action.

We now demonstrate how the Player Roles module helps to find a suitable replacement for Manchester City midfielder Fernandinho. Hence, we select the Brazilian midfielder as a reference player in this demo. Based on the characteristics of the players Manchester City acquired between the winter of 2018 and the winter of 2019, we set the age filter to a maximum age of 24 and the SciSkill filter to a minimum value of 95.

Figure 2: Overview of candidate replacements of Fernandinho at Manchester City

The figure above shows the candidate replacements for Fernandinho suggested by the SciSports platform based on our search query. The list is topped by Harry Winks (Tottenham Hotspur), Leandro Paredes (Paris Saint-Germain) and Adrien Rabiot (Paris-Saint Germain). The figures below show the player roles fingerprints for Fernandinho, Winks and Rabiot. The fingerprints present a complete overview of the roles a certain player has fulfilled in the past eight months and enable to quickly compare two players. In this example, Rabiot’s fingerprint reveals that he is more of a Box-To-Box Midfielder than Fernandinho.

Figure 3: Player role fingerprints for Fernandinho, Harry Winks and Adrien Rabiot

Conclusion

The Player Roles module extends the search capabilities of the SciSports platform. The newly released module enables to find players who are similar to a certain reference player in terms of playing style in an intuitive manner. The additional filter reduces the number of shortlisted players even further and enables scouts to spend more time on video analysis and watching players live in action.