Machine learning : football results prediction model

Abstract

Being the world’s most popular sports, predicting the results of Football games has always been an interest worldwide. With the increase in availability of football games statistics, using machine learning became more relevant and feasible. This study aims to use previous football matches statistics to predict the future result by applying machine learning on specific subsets of data. The hypothesis is that using machine learning we will be able to have a high prediction percentage given the correct features for the model. Using the convolution neural network machine learning model, we built multiple models to predict football results. We tested the models on a dataset gathered from eleven European countries and spanning across eight football seasons with our main focus being the English Premier League. The results showed that the more data features we were able to provide, the higher the prediction rate for football results. The obtained results showed that by using machine learning we can predict football games results by knowing the participating players in the future matches. These results can open up further research topics such as creating a model to assist managers in team selection for optimizing results.