Human Decision Making and Artificial Intelligence: A Comparison in the Domain of Sports Prediction

Artificial intelligence (AI) research has become prominent in both academia and industry. With this, an interest in AI's ability to make sound decisions when compared to human decision making has grown. Predicting the outcome of sporting events has traditionally been seen as a difficult task, due to the complex relationships between variables of interest. Attempts to make accurate predictions are fraught with biases owing to the bounded rationality within which human decision making functions. This study puts forward the position that an AI approach using machine learning will yield a comparable level of accuracy. A random forest classification algorithm was employed to predict match outcomes in the 2015 Rugby World Cup. The performance of this model was compared to aggregate results from Super-Bru and OddsPortal. The machine learning based system achieved an accuracy of 89.58% with 95%-CI (77.83, 95.47) vs. 85.42% with 95%-CI (72.83, 92.75) for the platforms. These results indicate that for rugby, over the limited period of a specific tournament, the evidence was not strong enough to suggest that a human agent is superior in terms of accuracy when predicting match outcomes compared to a machine learning approach, at a significance level α = 0.05. However, the model was better able to estimate probabilities as measured by monetary winnings from betting rounds compared to the two platforms.