June 29, 2018
Betting on AI in the 2018 World Cup
In 2014, a group of German statisticians correctly predicted their own country would prevail in the World Cup. This year, statisticians at the University of Dortmund are upping the ante incorporating AI predictive analytics into their betting strategy. An article recently published by PBS outlines how this works. As of Friday (June 22nd), the model correctly predicted the outcome of 15 out of 24 matches.
Their predicted winner this year? Spain.
Predicting World Cup results boils down to correctly guessing the number of goals scored in any given game. Once the number of expected goals is established, analysts then run simulations to determine the winners and losers. The results also hinge on which teams play each other in the knockout rounds. If the best teams face off before the final, it significantly reduces their chances of winning the tournament.
The World Cup offers a chance for corporate banks and academic institutions to test economic theory and statistical modeling in real time. The goal is to get the goals right. Determining the correct number of goals scored greatly improves the accuracy of the predicted outcomes.
The University of Dortmund team grouped the ability scores of the World Cup squads in a machine learning algorithm called “random forest”. The algorithm also takes into account other team characteristics including their country GDP, average age of the players, FIFA rankings (which aren’t highly relied upon) and bookmaker odds.
The algorithm then uses decision trees to separate the high and low scoring teams to predict which ones will advance. This is a common probability model used in economics. The only difference? The model makes these predictions by itself.
The team still doesn’t fully understand how the model computes its results. It takes into account a substantial amount of data arguably exceeding human comprehension. Plus, this is sports after all! Upsets in this World Cup may already skew the AI model’s predictions. If we could accurately predict the outcome of sporting events using AI every time, then why watch sports at all?