Prediction of Football Matches Results: Decision Forest against Neural Networks

Modelling football match outcomes is a popular prediction strategy based on either statistical approaches, machine learning approaches or rating system approaches. This paper is set to study football match outcomes prediction (win, lose and draw) for English Premier League Season 2005-2006 consisting of 20 football teams which played 380 matches per season in term of machine learning approaches. Two prediction algorithms are proposed, which are the Multiclass Neural Network and the Multiclass Decision Forest. Past research showed that both decision tree and neural network have capabilities to predict football matches outcomes. Although decision forest is a extended version of decision tree, there are few comparative study between neural network and decision tree in football matches outcomes prediction. The comparative experiments were carried using the Azure Machine Learning Studio and the results showed that the Multiclass Decision Forest performed the best as compared to Multiclass Neural Networks.