Creating a model for expected Goals in football using qualitative player information

The field of sports analytics has been growing a lot in recent years. Sports like baseball and basketball were among the first to embrace it, but football has also taken big steps in that direction. One of the causes is that data analysis allows for the development of new advanced metrics which can provide a competitive advantage. This project presents a new version of one of these advanced metrics applied to football, the Expected Goals. The metric estimates how likely it is for a shot to end up becoming a goal. We present two different approaches for building the predictors: one that uses player qualitative information and another player agnostic. We then reflect on the importance of the calibration of the probabilities yielded by the models, as well as their possible interpretations, and present some of the applications that can be used to evaluate team and player performance. We also show the impact each feature has on the models to make their outputs interpretable and to demonstrate that the addition of the player qualitative information is important for the performance of the model