Predicting Game Results for Football League Using Deep Learning

This research aims to examine the most effective methodology for predicting football match outcomes within a league format, specifically focusing on the English Premier League. The predictions are solely based on analyzing previous match data, without any information about the current game, such as player lineups or whether the match is played at the home or away stadium. We compared the prediction accuracy of two models, the Multilayer Perceptron and Convolutional Neural Networks, using a training dataset of 1,160 matches. The Convolutional Neural Networks, incorporating our player performance metrics, achieved the highest accuracy of 55.31% when tested with 320 matches, particularly in predicting home team wins with an accuracy of 71.52%. Notably, this accuracy surpasses that of football experts and bookmakers.