Comparative Analysis for Predicting Football Match Outcomes based on Poisson Models

Abstract

Association football is a team sport played between two teams of 11 players, with approximately 250 million players in over 200 countries and dependencies, making it the world’s most popular sport. Predicting the results of football matches has been chosen by so many people who love football. This study has three objectives which are to predict the football matches outcome using Poisson models, to predict football matches outcomes using Poisson Regression model and lastly is to evaluate the results using prediction accuracy, log-likelihood, AIC value and R-squared value. A comparative analysis on football match outcomes based three variations of Poisson models from the literature, which are the default model, Dixon-Coles model, Rue-Salvesen model, and Poisson Regression model have been runned. The experiments are conducted using the RStudio with the football match dataset taken from English Premier League Season 2015/2016 until 2020/2021 as the training dataset and Season 2021/2022 as the testing dataset. The experiments showed that the best prediction model is the Dixon-Coles Poisson model with the highest log-likelihood of -15023.29 and the lowest AIC value of 32056.57.