6. Scouting and evaluating actions

6.1 Plus/minus models

In this lecture I introduce different ways of scouting player actions using data.

References:

Kharrat, Tarak, Ian G. McHale, and Javier López Peña. "Plus–minus player ratings for soccer." European Journal of Operational Research 283, no. 2 (2020): 726-736. Links to an external site.

Hvattum, Lars Magnus. "A comprehensive review of plus-minus ratings for evaluating individual players in team sports." International Journal of Computer Science in Sport 18, no. 1 (2019): 1-23. Links to an external site.

McHale, Ian G., Philip A. Scarf, and David E. Folker. "On the development of a soccer player performance rating system for the English Premier League." Interfaces 42, no. 4 (2012): 339-351. Links to an external site.

Links to an external site.

6.2 Percentiles and player radars

6.3 Markov models

6.4 Possession chain models

 

6.5 Valuing actions: overview

Lotte Bransen gives an introduction to a series of lectures on valuing actions in football using statistical models and machine learning. 

  1. Why go beyond traditional statistics to assess football players?
  2. How to assess the performances of football players?
  3. What does the VAEP framework have to offer?

All the code for this tutorial series is available here: https://github.com/SciSports-Labs/fot.. Links to an external site..

6.6 Valuing actions: handling data

Jan Van Haaren talks us through the pipeline for evaluating players using an action value based approach. 

6.7 Valuing actions: generating features

Lotte Bransen talks us through the the process of generating features for evaluating actions. 

1, Investigate data in SPADL represenation.

2, Construct features to represent actions.

3, Construct features to represent game states.

GitHub repository: https://github.com/SciSports-Labs/fot Links to an external site....

6.8 Valuing actions: training a model

Jan Van Haaren talks us through the process of training various machine learning models on action data.

  1. Split the dataset into a training set and a test set.
  2. Construct the baseline classifiers by using conservative hyperparameters for the learning algorithm.
  3. Optimize the classifiers by tuning the hyperparameters for the learning algorithm.
  4. Construct the final classifiers using the optimal hyperparameters for the learning algorithm.

GitHub repository: https://github.com/SciSports-Labs/fot Links to an external site....

6.9 Valuing actions: analysing models and results

Lotte concludes the series, looking at how to interpret the results.

6.10 Discussion with Lotte and Jan. 

In this video Lotte gives an overview of the approach taken above and we discuss how these rankings work.

6.11 Player rankings

This video explains another, alternative way of ranking players, also based on machine learning.