Predicting European top 5 league football match results based on EA series football video game data and betting odds

This paper introduces a new dataset that utilizes EA football video game data to predict actual game outcomes. The dataset includes real match records from five seasons across Europe's five biggest leagues, as well as player ratings in seven categories (attack, skill, movement, power, mentality, defense, goalkeeping) and 34 skill scores. In addition, actual betting odds from various bookmakers are included for match predictions. By utilizing this dataset, the XGBoost, Random Forest, CNN, LSTM, and SVM models employed in our study outperformed the baseline model, Which composed of odds and achieving accuracy rates of 52.6% for match outcome prediction and 59.6% for over/under total 2.5 goals scored predictions. Among these models, Random Forest demonstrated the best performance, with accuracy rates of 55.9% and 62.7% for match outcome and over/under total 2.5 goals scored predictions, respectively, representing improvements of 6.3% and 5.2% over the baseline models.