There is not one specific metric that tells us whether a given player career advice is good or not. Therefore, we look at several other factors that influence the success of a transfer, such as the number of minutes played and the development of a player.
When we look at historical transfers, we know whether a player played a lot of minutes at his new club and whether he made a development (in terms of SciSkill) since his transfer. This is the information that we use to train the playing minutes prediction model and the development prediction model. Besides that, for historical transfers we also know the SciSkill, Potential, age, market value of the player at the time of the transfer and the fee paid. We use this information to train the transfer likeliness model.
After we evaluated the underlying models, we also did some tests to make sure that the club fit score given to a player-club combination made sense. For this, we used experts from within the football industry who provided us with information about the top 10 clubs that our model found for a selection of players. For each of those clubs, the experts told us whether it would be a good fit for the player and if not, why not. Using this feedback we could improve the models and for example decided that we should include recent SciSkill development to the models to provide better advice for players that made a big development in the last couple of months.