Predicting the Result of English Premier League Matches

Abstract

Nowadays predictive models or prediction of results in any sports become popular in data mining community, and particularly, English Premier League (EPL) in football gains way much attention in the past few years. There are three main approaches to predict the results: statistical approaches, machine learning approaches, and the Bayesian approaches. In this paper, the approach used is machine learning and evaluating all features that influences the results and attempts to choose the most significant features that lead a football team to win, lose, or draw and even considering the top teams. This predictive model basically helps in betting areas and also for managers to have a knowledge how to set up their team by analyzing the results also companies like StatsBomb which use these kinds of tools for setting up scouting networks for searching of hidden gems throughout the world. These features help predict the best possible outcome of the EPL matches using these classifiers logistic regression, support vector machine, random forest, and XGBoost; the data used for prediction is selected from the Web site: datahub.io, and the model is based on the data of last ten seasons of EPL. K-fold cross-validation is used to describe the accuracy of the model.