The SciSports Career Advice Application uses data-driven solutions to help players make the best-informed transfer decisions. For this, a range of metrics are used to ultimately find the best club fit within specific leagues of interest to the player.

Below, we take a detailed look at how each of the six metrics are computed.

Likeliness

Transfer likeliness is powered by a machine-learned model that examines the last 5 years of transfer activity for the club in question. For all those transfers it looks at the following attributes of the transferred player:

  • SciSkill

  • Potential

  • Age

  • Transfer fee paid

If we then want to estimate the transfer likeliness of player X to club Y, the model looks whether such transfers were done by this club before. For example, moving a younger player to a club traditionally focussed on experienced professionals may be more unlikely to happen. Or, club X never paid more than 5 million euros for a player then the transfer likeliness of Player Y with a market value of 10 million euros will be very low. Similarly, if a club always tends to buy younger players with a SciSkill within 60-80 and Potential between 100-120, then the transfer likelihood will be low for a player aged 30, SciSkill 90 and Potential 90.

Through this way of examining each club for the chosen player, users can have a quick overview of how likely the transfer is to happen from a historical perspective. Weighing this up alongside other metrics detailed below provides useful insights overall.

Playing time

Our Playing Time Predictor is a machine-learned model that predicts the share of minutes played by the player if he would make the move to the new club. Essentially, players and agents need to know whether or not the move would benefit them from a playing perspective. This metric ensures that there is a way of being able to predict possible playing time.

The model has been trained on 42,222 unique player-season combinations from 2016-2018 in a wide selection of leagues. For each player, we looked at the actual minutes played in this season and whether he is new to the team. The model takes the following things into account:

  • The player's SciSkill in relation to the SciSkill values of his (new) team mates

  • The player's Potential in relation to the SciSkill values of his (new) team mates

  • The player's age

  • The player's recent SciSkill development

  • The player's position

  • The number of players that can play in the same position line (e.g. Goalkeeper, Defender, Midfielder, Striker)

  • The number of players that have the exact same main position

  • The SciSkill values of the players that can play on the player's position

  • The SciSkill values of the players that have the exact same main position

When this model is run in the application for a given player and club, we only take into account the players that would still play at this club after the upcoming transfer window. Confirmed transfers are taken into account, transfer rumours are not taken into account.

For more about the evaluation of the model, please see this article.

Development

Our Potential Development Predictor is a machine-learned model that predicts the increase/decrease of the player's SciSkill in the next 12 months if the player would make the move to the new club. Essentially, players and agents need to know whether or not the move would benefit them from a development perspective.

The model has been trained on 34,388 unique player-season combinations from 2016-2018 in a wide selection of leagues. For each player we look at the actual SciSkill development made between the start of the season and the end of the season. The model takes the following things into account:

  • The player's SciSkill in relation to the SciSkill values of his (new) team mates

  • The player's Potential in relation to the SciSkill values of his (new) team mates

  • The player's age

  • The player's recent SciSkill development

  • The player's position

  • The SciSkill values of the players in the league

  • The SciSkill values of the top 11 players of all teams in the league

When this model is run in the application for a given player and club, we only take into account the players that would still play at this club and in this league after the transfer window. Confirmed transfers are taken into account, transfer rumours are not taken into account.

Playing style fit

This model looks at the Player Roles of the players at the club that played in the past two seasons on the position for which the model is run. For example, if we run the model for the position of left winger and we look at Club Y we look at all players who played as a left winger in the past two seasons. For example, it might be that this club always plays with a Classic Winger type of player as a left winger, then this club might not be a good fit for an Inside Forward.

Formation fit

This model looks at the formations that this club played in in the last two seasons. It especially looks at the position line for which the model is run. For example, if we run the model for a right back who is used to playing in a four-man defence, he would have a low formation fit with a club that always plays with five at the back.

Club status

This model looks at the status of the club which is based on the market values of the players they have in the squad.

Did this answer your question?