Machine learning can't predict World Cup winners because the tournament happens only every four years, giving you maybe twenty data points to train on. International football is too chaotic: low scores mean single goals change everything, red cards and VAR decisions are unpredictable, and one loss eliminates you. A prediction market works better than any model.
Let's build a machine learning model to predict the World Cup winner, except it won't work, and here's why. The World Cup happens once every 4 years, which means we have maybe 20 tournaments of reliable data, which is far from enough for a model to learn from. Club data would not fill this gap because international football has such a different dynamic. We have players who've barely trained together and much less history between the opponents. Football is also a very low-scoring game compared to other sports like basketball. There just aren't enough goals for the statistics to average out. A logistic regression model trained on 2-1 games is essentially just guessing randomly. One goal at the last minute can change the entire outcome, and we've seen that time and time again. Human judgment is also a huge component of World Cup games. From an unexpected red card followed by an even more unexpected reversal to the VAR controversy that led to Argentina's win over Egypt. There are just so many examples of important judgment calls that the model cannot capture. Then there's also the format itself. In a league season, you have 38 games where the good ones can cancel out the bad ones. In the World Cup, you have one bad game and you're going home. Now you could try to update the model parameters in real time as the games go on, but at this point you're still trying to model the team chemistry, the motivation, the injuries, and potential chaos, with only a handful of data points. Updating the model doesn't change the fact that there's barely a model to begin with. So in this case, compared to machine learning, a real-time updated prediction market might be more efficient. But at this point, I'm probably trusting the seal more than anyone, and not just because I want Rory to win tomorrow.
