A case study of Identifying Low Blocks and Strategies in Football with FC Nordsjælland
Football, like various team sports, is an invasion game where opponents incorporate collective strategies to attack each other's territories. Most popularly, football has three moments:
1) Attack (In-Possession)
2) Defence (Out-of-Possession)
3) Transition (Ball is being disputed in the transition to (1) or (2))
This study focused on 1) and 2) interchangeably and strategies used by teams when in those moments.
Most recently, I had a great opportunity to present a novel approach to identifying low-block scenarios and effective strategies used against a team well-positioned in a low-block. I presented these findings at the Opta 2020 Forum. So, what is a low-block? A common term like “Parking-the-bus” has floated around but to define it more aptly:
“ The low block is a defensive tactic in which a team defends from a very deep position on the pitch, meaning they then restrict the amount of space the opposition has to exploit. As a result, opposing teams tend to find it much more difficult to score due to the lack of room to operate in central areas of the pitch” — Football Scotland
This study was based on a club-led research question posed by FC Nordsjaelland, a Danish Super Liga Club. The club is directly linked to the ‘Right to Dream Academy’ (RDA) in Ghana, instrumental in recruiting young football talent. The strategic objective of the club is to spend less money on recruitment at a senior level whilst giving opportunities to young talent. Over the past 12 years, The Right to Dream Academy has produced talented individuals who have taken up scholarships worth $45 million through the US collegiate system.
Defining the problem: Real-world scenario
The head coach of FC Nordjallend, Flemming Pedersen, idolizes football with a pro-active possession-based approach. During the winter months, teams who play against FC Nordjallend most often defend deeper towards their own goal. The head coach seeks to identify and evaluate strategies that are influential to the game model when FC Nordjallend face oppositions well positioned in a low-block scenario.
“We have a gut feel for how effective our strategies are at breaking a low-block, but we have no hard evidence to challenge our assumptions. Using data to evaluate how we implement our game model is its most interesting use case” — Flemming Pedersen, Head Coach
Optical Tracking & Event Data
Event data primarily captures on-the-ball actions across the pitch and is supplied by Opta-Stats Perform. It predominantly covers all the actions in a game with respect to the ball location. The optical tracking data is positional data of players and the ball captured during a game at 25 frames per second. Optical tracking data is crucial to identify the movement of players on the football pitch which is clearly illustrated by the famous Johan Cruyff:
”When you play a match, it is statistically proven that players actually have the ball 3 minutes on average … So, the most important thing is: what do you do during those 87 minutes when you do not have the ball. That is what determines whether you’re a good player or not.” — Johan Cruyff
This project used a combination of optical tracking and event data.
Temporally geometrical centroids
Temporally geometrical centroids were identified as an intuitive approach to identify teams’ positions when faced with an attack by the opposition. The premise here suggests that if a team is centrally located deep towards their own goal, their combined centroid locations would give a better 1-dimensional representation. Ideally, using a point dimension to represent the location of a team is intuitive and can be used as a preliminary assumption to add on further rules to identify a low-block region.
Identifying when the team has broken a low-block
The next step was to identify when these low-block situations were broken by the opposition.
behind the last two defenders
Capturing Offensive Strategies against a Low-Block
Once the low block was identified against the opposition, the project looked at identifying strategies used by FC Nordjallend against these situations. The strategies were broadly defined by coaches which were translated through data.
The Strategies used were:
- Creating overloads in wide areas of the pitch: Encourage players to draw in defenders and analyse the right moment to break the opposition with runs into the half-space
2. Wide Pocket to Half-Space Run: Possession in wide areas, identifying gaps in defence within the half-space and making assertive runs behind the defence into the half-space
3. Fast Switches to Wide Zones: Importance of moving the ball quicker into wide areas when central areas are overloaded
4. Decisive off-ball runs at the defensive line: Assertive runs to be attempted when a ball is played behind the defensive line
5. Receiving in central pockets: Players are to receive in central pockets with the ability to play forwards. Numerous run patterns towards the defensive line by other players are made simultaneously.
Putting it together to analyse results
Key Takeaways
This project provides a framework for using a rules-based approach to then build machine learning models. This strategy could be easily incorporated in pre-match and post-match analysis. More importantly, due to the effectiveness of the approach, performance analysts can view similar scenarios during live games and inform coaches sitting in the dugout.
Similar effective approaches are used by data teams at FC Barcelona and Toronto FC amongst other elite clubs. This project embraced coach recommendations through an iterative process that helps achieve the quintessential “coach-buy in” and thus challenges coaching practise. This is crucial because it bridges the gap between the two worlds of football and analytics.
Future work
Long term projects such as complex machine learning models could help add value but are compute-intensive. Football is a nuanced sport which is constantly evolving and any breakthrough that adds value to outcomes would be well-received. With that in mind, some of the ideas generated may include:
- Using optimization models to minimize or maximise the time taken for a runner to arrive at a certain moment. For instance, playing to team strengths: Can a player be located in a position where he can create maximum value based on his strengths at a specific time?
- Predicting successive time sequences using a neural network where a low-block is broken
- Sensitivity analysis to understand key metrics that influence breaking the lines/breaking a low-block
- Identifying similar approaches to identify open-spaces in football or options for players to take up spaces during build-up play. This could help dictate effective attacking patterns of play
- Using Transfer learning approaches such as vacant parking-lot spaces to identify empty spaces on a pitch
Finally, I would like to thank Joe Mulberry who has been of immense help working on this project. If anyone would like to discuss ideas, questions or collaborate on any projects kindly get in touch on Twitter or LinkedIn!
Thank you!
