Academic Journal - Applied Intelligence - 15 November 2023
Mahdi Nouraie, Changiz Eslahchi, Arnold Baca3
https://link.springer.com/content/pdf/10.1007/s10489-023-05150-x.pdf
In this paper they propose using a data driven approach called Intelligent Team Formation & Player selection to help choose which players to play and in what formation would work best. They combined how well players are rated to suit different potions against the graph theory to decide on the optimal lineup which will produce the best outcomes for the team. They tested these theories with 4 premier league clubs during the 2021–2022 season and compared the data driven decisions made against the decisions with actual coaches’ choices.
Due to football coaches normally relying on experience and intuition to choose the formation and line ups for a team, this data driven system helps coaches to make smarter and more objective choices.
The goal of it is to match players to positions based on their performance and positional suitability on the pitch. In this paper, this was done using a two stage framework. Satfe one looked at how suitable each player is for each on field position. This looked at statistics like the players’ technical, tactical and physical attributes and the system outputted a positional suitability score for each player-postion pair. The second stage was then using the Hungarian algorithm to assign players to positions for the maximum total suitability score, resulting in the vest starting 11 with all players in their optimal positions.
The outcome of this test was that the data driven approach closely matched what professional coaches had also picked for top teams such as Manchester City and Liverpool, with an on average 80% agreement. The outcomes that differed were usually down to contextual or tactical factors such as injuries or fatigue.
<aside> <img src="/icons/graduate_pink.svg" alt="/icons/graduate_pink.svg" width="40px" />
This data driven model would work the best for the pre match analytics as before a match a starting eleven and impactful substitutes need selected, coaches must decide on the formation and tactics and adjust each players role based on their opponents style and team balance.
This model is perfect for this as its use case is primarily for planning lineups and formations based on the data. It provides coaches with evidence based support for their line up and formation choices as they can input different formations to see how the layer suitability and overall team score would range, highlighting which formation provides the best efficiency for the squads attributes
However as this model uses FIFA data (technical and physical player attributes) rather than match specific form or contextual data (current fitness, recent performance), how a player plays on the day under the conditions may be different to what was expected of them. It also assumes that the scoring can capture the complex data of real tactical interaction but in reality it wont understand intangible factors such as chemistry, recent form or psychological readiness.
This model also fails to look at opponents behaviour and how it may impact the game. A coaches optimal formation should reflect not only player strengths but also how they match up tactically against the opponent.
</aside>