KU Leuven lab reveals tactical value of kicking ball out of play
The Sports Analytics Lab at KU Leuven has utilised machine learning to quantify the effectiveness of counterintuitive soccer tactics, while advancing tools to automate game data annotation.
Jesse Davis, a computer science professor at KU Leuven and head of its Sports Analytics Lab, has significantly influenced professional soccer tactics through the development and open-sourcing of machine learning tools. The lab’s research, including a 2024 paper titled “Boot it,” utilised data from over 1.4 million passes to demonstrate that kicking the ball out of bounds near the opponent’s goal can strategically reduce the actions required to score. The lab provides widely used frameworks such as VAEP and expected goals (xG) models, helping clubs evaluate player performance and tactical efficiency. Davis is also collaborating with other institutions to standardise in-game data using transformer neural networks to reduce manual annotation efforts.
The 2024 paper “Boot it” utilised data from over 1.4 million passes, partly drawn from the 2022 World Cup, to simulate the tactic using tree ensemble models. The specific finding suggests that kicking the ball out of bounds in the middle third of the pitch can put a team within 10 actions of a goal. Davis explains that the strategy involves setting up a recovery of the ball in an advantageous situation, a move that has appeared in top leagues despite appearing counterproductive to casual observers.
Davis’s lab provides widely used open-source frameworks, including VAEP (assessing action effects) and expected goals (xG) models, which help clubs evaluate player performance and tactical efficiency. These tools allow teams to quantify probabilities and assess the efficiency of strategies, such as the value of long shots. A 2021 study by postdoc Maaike Van Roy and colleagues, presented at the MIT Sloan Sports Analytics Conference, showed that Chelsea could gain 1.6 more goals per season by shooting from distance 20% more often.
The lab’s influence extends to clubs such as Royal Sporting Club Anderlecht and Club Brugge KV, where analysts apply the research to roster evaluation and tactical planning. Hugo Rios-Neto, data recruitment lead for Anderlecht, described Davis’s lab as the most influential in soccer, noting that the work helps uncover hidden tactical patterns. Van Haaren, a former lab member now at Club Brugge, stated that the collaboration centres on translating football philosophy into measurable, data-driven outputs.
Davis is collaborating with other institutions to standardise in-game data using transformer neural networks to reduce manual annotation efforts. The lab is experimenting with transformer neural networks (the architecture underpinning large language models) to train models to automatically tag tactics, such as three-on-two breakaways, based on limited human tagging. This effort aims to address the manual process of game annotation, which Davis described as a nightmare for data analysts due to its time-consuming nature.
