Improving the Expected Goal Value in Football Using Multilayer Perceptron Networks

The development of big data and machine learning is having a special boost in the field of innovation in sports. Football is the most popular sport in Europe with millions of players and billions of euros invested. Currently, machine learning applications in football are focused on video analysis to events detection (tracking of players, statistics matches, scouting, ...), injuries evaluation and prediction, among others. The xG metric (Expected Goals) determines the probability that a shot will result in a goal and is often displayed on screen during the most important football matches in the world. In this paper we present a new model to obtain more accurate predictions of the probability of a shot becoming a goal. The model is based on a multi-layer perceptron neural network (MLP), which allows the interaction of different variables and improves the model’s performance. Our proposal includes an evaluation of the quality and a comparison with the xG provided by statsbomb, one of the most important football data providers of the world. The results show that our model outperforms the quality of the expectations provided by statsbomb. Specifically, it is clearly better in their capability to detect actual positive cases (goals).