Finding ways to improve the prediction accuracy of a model that predicts the outcome of a football match using machine learning

The goal of this research is to find new ways to improve the prediction accuracy of a model that predicts the outcome of a football match. In our literature review, we found out that there has been quite some research focusing on finding new and effective feature categories or algorithms. Most of these would make use of the known effective feature categories, like match attributes, match statistics and team performance, while introducing a new feature category. Also, some of these would experiment with a new algorithm to find out whether it would have potential. Studies like [1], [2] and [3] made use of these ways to improve the prediction accuracy and have had some success over the years. [1] made use of weather as a new feature category, while [2] focused on team/player ratings and team/player values. [3] did not focus on a new feature category but used the known effective ones, while experimenting with long short-term memory. As mentioned, these ways to improve the accuracy have had their success over the years but at a certain point the pile with new feature categories and algorithms will run out. In that case, we need to find other ways to improve the prediction accuracy of a model that predicts the outcome of a football match. In this thesis, we propose two ways to improve the prediction accuracy other than finding new feature categories or algorithms