Research Article - 14 July 2020

Pavlina Kröckel & Freimut Bodendorf

https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2020.00047/full

Article Aim

The aim of this article was to introduce process mining techniques in football anayltics and how these techniques could produce tactical insights to help analyse sequences of actions, player involvement, and event dependencies. It uses two contrasting teams England and Iceland in a Euro 2016 match, to show how process mining reveals differences in style, tactics and adaptability.

These sequences can help for pre-match preparation, in-match adjustments, and post-match evaluation as it highlights how using clustering can help to identify typical and exceptional playing sequences and behaviors

What The Article Covers

Process Mining

Process mining used to analyse and improve process by looking at events such as records of what happened, when, and who was involved. It sits between machine learning and process modelling. For process mining algorithm to work t will need to be structured with the following data: Case ID (a match or possession), Activity (pass, shot) and a Timestamp.

There is 3 types of process mining:

  1. Discovery is the most common. It builds a model from the data and shows what really happened.
  2. Conformance Checking compares the actual process to an exisiting model and finds differences or rule violations.
  3. Enhancement uses data to improve an existing model.

In football, only discovery is useful as there’s no fixed model for how a game should unfold.

Different Perspectives in Process Mining

  1. Control-Flow Perspective (How things happen)

    Shows the order of actions and patterns (like loops or repetitions in the game)

  2. Organisational Perspective (Who does what)

    Focuses on people or players and how they interact and uses metrics used such as who passes the action to who and who appears in the same sequence together

  3. Case Perspective (What happens in each scenario)

    It focuses on each unique case such as an attack or possession sequence and then uses tools such as trace or sequence clustering to help find common or unusual plays or patterns