• artificial intelligence
• data analysis
• rating prediction
• soccer analysis

Humans are routinely asked to evaluate the performance of other individuals, separating success from failure and affecting outcomes from science to sports. In this thesis, we focus on the case of performance evaluation in soccer. Specifically, we first create a dataset capturing performance evaluation of soccer players by sports journalists. We combine these ratings with a massive data set of spatio-temporal events describing the behavior of players during each match of the Italian first division. We then ask the following question: can we reproduce, using Artificial Intelligence, the way journalists rate soccer performance? To address this question, we train a set of machine learning models that predict a player's rating in a match given their performance as described by a vector of performance features. We find that our models can reproduce the human evaluation process, a result confirmed by a set of experiments conducted with human evaluators. Second, we use model interpretation techniques to investigate the reasoning behind the models' predictions, providing a way to explain what are the key factors human evaluators focus, as well as a tool to explain why a player performed well, or bad, during a specific match.