This research is a continuation of Baird (2002), and the research is also focuses on probabilistic sports prediction. In probabilistic sports prediction, the participants are required to predict the probability of a team winning a match during the Australian Football League (AFL) season. Machine learning methods such as Maximum Likelihood (ML), Bayesian Model Averaging (BMA) and Minimum Message Length (MML) are used to estimate the probabilities of winning.

To estimate the probability, the estimators are applied to three models: "weight", "boldness" and joint model of the previous two.

Three simulations are performed. The first two aims to compare the performance of the machine learning methods by training them in both limited and extended data case. The third simulation compares the performace of the A.I. tipper (the 3 estimators) with the "real" tippers from the Probabilistic Football Competition run by the CSSE.

Download