Robo2018 hangs out a lot with machines. So much so, that he has extended the friendship to the soccer World Cup underway in Russia.
No, ‘Robo2018’ is not an android. He’s very much human, a statistics graduate who goes by the day name of Alexander Vosseler. He has a day job too. Being a Senior Data Scientist in the Global Claims Analytics department of Allianz Global Corporate & Specialty (AGCS) has its advantages, especially when it comes to betting on the Russia World Cup.
Using the power of data, Vosseler predicts that France will lift the World Cup. He’s hoping to win the "Golden Boot" at the Allianz Prediction Game with Machine Learning, which is an internal prediction game for football-enthusiastic employees.
You’d think that the man who has made it to the second place in the World Cup Prediction Game after the preliminary round would be a football expert. Ironically, Vosseler admits he doesn’t know “anything about football”.
“In my work, I develop machine learning models with which we can, for example, automatically detect fraud cases, calculate the probability of successful recourse from third parties and much more. In principle, I used the same algorithms to predict football matches,” he says.
Vosseler explains how the model works: “In a first step, we collected all freely available data that was relevant to us. This, of course, includes all results of the last big tournaments, as well as the friendly matches of the last 10 years.” There on, he created variables to better predict the chances of a win or a draw. Among these variables were the current World Football Elo rating per team, goal differences per match and so on.
Two models that are frequently used in claims were combined to get the predictions. One was a model for predicting the "win/draw/loss" conditions and the second was for predicting goals per team. “Then I waited for ‘Robo2018’ to spit it out.”
Vosseler himself was rather surprised at how reliable the predictions were. “Of course, the machine is not clairvoyant. Here, correlations and patterns are learned on the basis of historical data and made usable for predictions.”
Unfortunately, the robot got Germany’s early ouster wrong. But that’s not surprising. Historical data had nothing to suggest the world champions would face a disappointment. “If Germany had never lost to South Korea and had never been eliminated in the preliminary round, the machine has a hard time predicting this in the end.”
Vosseler’s friends and family members are slowly warming up to his machine friends. “My father asked me to make some predictions the other day,” he laughs. “He has increased his private betting game by six ranks within two days.”
Pleased with the experiment, Vosseler is wondering if he should refine the model for the next Bundesliga season. “At the national championship, predictions should be much more accurate because there is much more data that we can feed the machines and you can expect much better hit rates if you can consider additional variables such as ‘home away game’. That's only possible to a limited extent at the World Cup.”
For now, Robo2018 is the cheerleader for France. “Obviously, being in the semi-finals means you always have a good chance of winning. However, this World Cup has been full of surprises and unexpected outcomes already, so we wouldn’t bet on our robot’s predictions.”
Machines rooting for men. How’s that for an exciting tournament?
As with all content published on this site, these statements are subject to our Forward Looking Statement disclaimer: