Data Analysis

Tech how-to: Build your own expected-goals model

09/05/18 | SciSports

Football is a low-scoring game, which means that goals are usually a poor indicator of the actual performances of the players and the teams on the pitch. The observation that shots occur more frequently than goals has inspired football analytics researchers to develop several shot-based performance metrics. Undoubtedly, the most widely-adopted shot-based performance metric in football is the expected-goals value for a shot, which reflects the shot's likelihood of resulting in a goal based on the information that was available prior to the shot. For example, a shot that has an expected-goals value of 0.15 has a 15% chance of resulting in a goal.

In this tech how-to, we guide you through the process of building a simple expected-goals model that produces an expected-goals value for each shot that happened during a game of football. Our how-to is available as a Jupyter notebook and showcases some of the tools and technologies our data analytics team uses to build their analytics models and metrics. In addition, we release an artifical but realistic shots database containing information on 127,643 shots to stimulate football analytics research.

Want to build your own expected-goals model for football? Check out our tech how-to to learn more!