Predicting the Outcomes of Football Matches Using Machine Learning Approach

Predicting outcomes of football matches is among the rapid growing area of research due to the interest of large number of people, and the stochastic nature of the results. Many researches have been conducted to predict the outcomes of football matches. Statisticians predict the football match outcomes to showcase their skills whereas Operational researchers use the prediction to experiment with various effects of football tournament design. This research is aimed at determining feature and classifier combination that would provide the best English premier League football outcomes prediction accuracy. Despite the fact that feature and classifier combination might effects the football outcomes prediction accuracy, however, based on our knowledge, no other researches had considered to determine which possible feature and classifier would produce a good English Premier League football prediction accuracy. We proposed to use multiple machine learning classifiers and features combination to determine which feature and classifier combination would produce high football prediction accuracy. Experiments were conducted using various feature and classifier combinations including previous football match records of English premier league for two seasons. The results of all feature and classifier combinations were compared to determine the one that would achieve the highest accuracy. The result of the experiments show that using home team factor, away team, twenty two (22) first players from the two teams, and goal difference together with K-NN machine learning classifier achieved the highest accuracy of 83.95%. In addition, the result of the best feature and classifier combination was compared with the result of the existing English Premier League football prediction accuracy and found that our result achieved improvement over existing results.