Continue to the content
The best Red Devil: the man of the match via machine learning
Research

The best Red Devil: the man of the match via machine learning

Researchers at KU Leuven are bringing scientific rigour to the sport by building machine learning models.

5 minutes
06 January 2020

If avid football fans can no longer cheer on their favourite teams and players in the stadium, then they’ll gladly do so from home, in front of the TV set. One of the reasons football is the world’s most popular sport is its inherent unpredictability. For one, the underdog can beat the hot favourite because of a single player’s split decision. Moreover, whether the ball hits the post or just sneaks over the line can hugely impact the outcome of a match. Even though football is considered one of the hardest sports to analyse, researchers at KU Leuven and data intelligence company SciSports are bringing scientific rigour to the sport by building machine learning models. Owing to the models’ thorough analysis and predictions, we are gradually moving closer to understanding the beautiful game of football a little bit better.

How Artificial Intelligence Developed at KU Leuven is Raising the Football Game

We have all witnessed – or taken part – in the traditional post-match debates about which footballer performed best or worst on the pitch that day. But what if science could actually track these variables and settle the debate over who is best once and for all? Rather than counting on traditional player statistics like the number of goals and assists, a new model designed by researchers from the Department of Computer Science at KU Leuven and Dutch data intelligence company SciSports provides us with a more comprehensive assessment of a player’s performance.

Goat voetbal duel

Eye on the ball

KU Leuven Professor Jesse Davis and his team collaborated with SciSports to develop an algorithm, which can value a player’s total impact on the game. More than counting goals and assists and generating traditional player statistics, the computer model takes into account a player’s every single on-the-ball action, ranging from passes and shots to tackles and dribbles.

The rating this model generates is called ‘VAEP’, or ‘Valuing Actions by Estimating Probabilities’. The VAEP rating shows us how an individual football player’s actions contribute to the final score. This rating considers both offensive moves, such as scoring goals, as well as defensive conduct, i.e. preventing the opposition from scoring goals, allowing a detailed understanding of the players’ behaviour on the pitch. This way, we end up with a much more complete analysis of the footballer’s achievements and his overall contribution to the team.

This seamlessly brings us to what is, at the moment, one of the most hotly debated topics among football aficionados: the position of Portuguese superstar Cristiano Ronaldo versus Argentine football hero Lionel Messi. Based on the algorithm, KU Leuven PhD student Tom Decroos can conclude that, although both players have near flawless track records, and have been awarded the prestigious Ballon d’Or no fewer than five times, Messi has a slight edge over Ronaldo. In fact, Decroos’ analysis was solidified when, after yet another scintillating season, Barcelona forward Messi clinched his sixth Ballon d'Or in 2019.

Scouting skills

Although thus far the results of VAEP framework are promising, Professor Davis and PhD student Decroos recognise there is still room for improvement. As soon as the researchers have access to data of footballers’ actions that are not directly on the ball, an even more fine-grained analysis will be possible.

Data analytics have become increasingly important for scouting and player acquisition. The current version of the VAEP framework is already highly useful for seeking out fresh new talent and assessing their true potential. Moreover, Professor Davis’ research collaboration with SciSports is creating a buzz far beyond national borders. In fact, their study was selected out of seven hundred submissions at the highly acclaimed KDD conference in Anchorage, Alaska to win the Best Paper Award for applied data science.

Mind over body?

However tense we are when watching a football player getting ready to take a game-winning (or –losing) penalty kick, does it even compare to what the player himself is experiencing? Football commentators and fans usually focus on the sport’s physical aspect, the underlying tactics and the technicality of the performances, whereas little to no attention is paid to the mental aspect of the game. KU Leuven and SciSports are now changing this. They were, in fact, the first ones to look at how well players perform under pressure. Their findings underline how professional football is as much a physical game as it is a mental endeavour.

Goat decision

Professor Davis and his team designed a machine learning-based model that is able to gauge the amount of pressure under which the player who has possession of the ball finds himself. By analysing a whopping 7,000 matches in seven competitions, the researchers attempted to uncover which football player rises to the occasion, and who crumbles under pressure. The study takes various parameters into account in order to calculate how much stress players experience per minute both leading up to and during the match. Those levels are compared to a player's performance under normal circumstances. In addition, the model considers the decisions the player makes on the pitch and how well his actions are executed.

Make-or-break

In 2019, the study was presented at the MIT Sloan Sports Analytics Conference in Boston. Apart from being an exciting contribution to the sports analytics discussion, this KU Leuven and SciSports-led research can provide trainers with some invaluable insights; they could, for instance, consider how a player responds to high-stakes situations when composing a team for a make-or-break match, or even decide to teach more stress-prone players how to cope with high pressure-situations.

Goat bernabeu

Clearly, the groundbreaking research endeavours by Professor Davis and his team allow us to analyse football players and the game a lot more thoroughly; the many factors and variables that are taken into account by their models offer never before seen insights into the sport. However, as detailed an analysis their models might generate, the outcome of a football match will never be completely predictable – which is a good thing! Indeed, the immeasurable power, the ‘magic’ of this global sport, is what has been keeping spectators on the edge of their seat for the entirety of the game, and will continue to do so.