Expanding the Scope of Football Analytics by Integrating Tracking Data and Utilizing Statistical Learning

Abstract

The recent decade has seen a data revolution within football as data analytics and statistical learning have become established vital tools. The data revolution has also seen the emergence of a new type of football data called tracking data. This thesis first explores how information from tracking data can be integrated with established event data and improve an established statistical learning model providing an expectancy metric for passes called xP. Secondly, the thesis explores how it can used to create a new type of statistical expectancy metric for player playability previously unattainable with only event data. Using event data and tracking data from 28 real football games, these separate datasets have been synchronized to extract new information and context for passing events. This information was used to train and compare a statistical learning model for the xP metric with a model only trained on the previously known event data. The results indicate that the added tracking data information provides a significantly improved xP model especially in terms of understanding passing events and therefore making more realistic pass probability predictions. Despite clear improvement, there exist possibilities to further improve the xP model in regards to for example a more accurate data synchronization process as well as further improved feature engineering. Moreover, the synchronized tracking and event data in combination with the improved xP model were used to develop a metric that describe player playability expectancy called xPlay. The new metric provides a simple and elegant way of measuring player playability and results of various implementations indicate that the metric can serve as a great tool in both player and team evaluation. Although promising results the metric is in need of more evaluation on a bigger scale.