Glicko-2 model adapted for soccer.

Based on the paper by Andrei Shelopugin and Alexander Sirotkin titled 'Ratings of European and South American Football Leagues Based on Glicko-2 with Modifications'.

Abstract

This model endeavors to compute club ratings of the first and second European and South American leagues. In order to calculate these ratings, the authors have designed the Glicko-2 rating system based approach, which overcomes some Glicko-2 limitations. Particularly, the authors took into consideration the probability of the draw, the home-field advantage, and the property of teams to become stronger or weaker following their league transitions. Furthermore, authors have constructed a predictive model for forecasting match results based on the number of goals scored in previous matches. The metrics utilized in the analysis reveal that the Glicko-2 based approach exhibits a marginally superior level of accuracy when compared to the commonly used Poisson regression based approach. In addition, Glicko-2 based ratings offer greater interpretability and can find application in various analytics tasks, such as predicting soccer players metrics for forthcoming seasons or the detailed analysis of a players performance in preceding matches.

Introduction

You can find useful examples in jupyter notebooks:

QuickStart.ipynb: A gentle introduction to the Glicko-2 rating system.

Boostings.ipynb: Examples of how to train and use LightGBM and CatBoost models for predicting soccer match outcomes.

Ratings.ipynb: How to use our approach to calculate club and league ratings.

Citation

If you find this model helpful, please cite our work.

@misc{shelopugin2023ratings,
      title={Ratings of European and South American Football Leagues Based on Glicko-2 with Modifications}, 
      author={Andrei Shelopugin and Alexander Sirotkin},
      year={2023},
      eprint={2310.11459},
      archivePrefix={arXiv},
      primaryClass={stat.AP}
}