Preview
Hüseyin Akbulut, MSc (2026). The Future of Football Analytics: Beyond GPS to Spatial Intelligence. Sporeus. Retrieved, November 22, 2026. https://sporeus.com/en/sport/future-football-analytics-spatial-intelligence/
Introduction
The era of football analytics built on GPS data and event counts (distances, speeds, pass completion) is giving way to a more sophisticated approach: spatial intelligence. Every moment of a match is described not only by which actions occurred but also by the state of the game, which options were available and how much value was created or lost. This shift, from measuring outcomes to valuing actions in context, is a fundamental change in how performance is understood, and it is being driven by machine learning applied to tracking data.
Table of Contents
The Science
The limits of event-based statistics: traditional football statistics record outcomes: goals, assists, shots on target, pass completion percentage. These are noisy signals, heavily influenced by teammates, opponents and chance. A central midfielder who attempts ten ambitious passes into dangerous areas might complete only a few of them and look statistically worse than a player who makes forty short, safe passes at 95% completion. Event-based statistics systematically undervalue risk-taking and space creation.
From counting to valuing actions: models such as VAEP (Valuing Actions by Estimating Probabilities) and Expected Possession Value (EPV) estimate, for each game state, the probability that the possession will lead to a goal (and, in VAEP, also to conceding one). Every action, successful or not, changes the game state, and the value of an action is the change in those probabilities. This framework:
- Values a wide range of on-ball actions, including defensive ones such as interceptions and clearances
- Accounts for the risk and reward of attempted actions, not only completed ones
- Makes it easier to compare players who play in very different tactical contexts
Machine learning on tracking data: optical tracking systems record the positions of all 22 players and the ball many times per second (commonly around 25 frames per second). Such data are hard to model with traditional statistical methods. Approaches used in research include:
- Graph neural networks (GNNs): players are treated as nodes in a dynamic graph whose connections change with their movements, which allows team structure and space creation to be modelled
- Transformer and other sequence models: attention mechanisms that capture longer-range patterns in player movement across possessions
- Generative models: models that generate plausible player movements or match scenarios from historical data, allowing tactical simulation
Counterfactual analysis: one of the most interesting frontiers asks, “What would have happened if the player had made a different decision?” By comparing the action actually taken with the estimated value of the alternatives, analysts can ground coaching conversations in data: “In this possession you had three passing options; here is the estimated value of each, and here is why holding the ball cost us.”
Player similarity and recruitment: movement profiles derived from tracking data can be used to find players in other leagues who move in similar ways to a target player, which may say more about tactical fit than similarity based on event statistics alone.
What Research Says
Fernández, Bornn and Cervone (2019), presented at the MIT Sloan Sports Analytics Conference, introduced an expected possession value framework for football that uses deep learning components on tracking data to estimate the value of the game state continuously during a possession.
Decroos and colleagues (2019) introduced VAEP at the ACM SIGKDD conference, a framework that values every on-ball action by how it changes the probability of scoring and conceding in the near future, and released the associated tools publicly.
Bauer and Anzer (2021) developed a data-driven method in Data Mining and Knowledge Discovery that uses tracking and event data to detect counterpressing situations automatically and assess their success.
Pappalardo and colleagues (2019) published a large open dataset of match events from top competitions in Scientific Data, giving researchers a shared resource for developing and comparing models.
Did You Know? Much of the progress in football analytics has come from open resources: public event datasets, open-source libraries for action valuation and pitch control, and openly published conference papers. Clubs without large analytics departments can therefore explore many of these methods before investing in commercial products.
Applied to Football
Applying next-generation analytics:
- Add action-value metrics to player evaluation. VAEP- and EPV-style metrics are increasingly available; use them alongside traditional statistics rather than instead of them.
- Use counterfactual examples in coaching. Showing a player the options available in a specific possession can be more actionable than descriptive event summaries alone.
- Analyse team structure as a network. Models of player positions and movements show how a team creates or denies space and allow systematic comparison against different opponents.
- Use movement-based similarity in recruitment. Tracking-based profiles can complement event statistics when judging tactical fit.
- Invest in data infrastructure. The gap between clubs with and without full-match tracking data is likely to grow as these methods mature.
The Short Version
- Event counts are noisy outcome measures; action-value models give a richer picture of contribution
- VAEP and EPV estimate the value of actions in context, including actions invisible in traditional statistics
- Machine learning methods such as graph neural networks make it possible to model the complex spatial and temporal structure of tracking data
- Counterfactual analysis supports coaching conversations about decision quality, not just outcomes
- Tracking data infrastructure is the foundation for advanced football analytics
References
- Fernández, J., Bornn, L., & Cervone, D. (2019). Decomposing the immeasurable sport: a deep learning expected possession value framework for soccer. MIT Sloan Sports Analytics Conference 2019.
- Decroos, T., Bransen, L., Van Haaren, J., & Davis, J. (2019). Actions speak louder than goals: valuing player actions in soccer. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 1851–1861.
- Bauer, P., & Anzer, G. (2021). Data-driven detection of counterpressing in professional football. Data Mining and Knowledge Discovery, 35(5), 2009–2049.
- Pappalardo, L., Cintia, P., Rossi, A., Massucco, E., Ferragina, P., Pedreschi, D., & Giannotti, F. (2019). A public data set of spatio-temporal match events in soccer competitions. Scientific Data, 6(1), 236.
Hüseyin Akbulut is the founder of Sporeus and author of THRESHOLD (EŞİK), a 540-page Turkish-language book on endurance science.
- 540pp THRESHOLD Book
- MSc Sport Sciences
- Marmara University
Introduction
The era of football analytics built on GPS data and event counts (distances, speeds, pass completion) is giving way to a more sophisticated approach: spatial intelligence. Every moment of a match is described not only by which actions occurred but also by the state of…
The Science
The limits of event-based statistics: traditional football statistics record outcomes: goals, assists, shots on target, pass completion percentage. These are noisy signals, heavily influenced by teammates, opponents and chance. A central midfielder who attempts ten ambitious passes into dangerous areas might complete only a few…
What Research Says
Fernández, Bornn and Cervone (2019), presented at the MIT Sloan Sports Analytics Conference, introduced an expected possession value framework for football that uses deep learning components on tracking data to estimate the value of the game state continuously during a possession.
Applied to Football
Applying next-generation analytics: