Predictive Analytics and the Super Bowl

Did you know that the Kansas City Chiefs only had a 3.9% chance of winning the Super Bowl LIV title? 

Sports and Statistics

Sports and statistics have a long tradition. Since the 1990s, statisticians have been teaming up with different sports leagues to improve game outcomes, such as predicting which team is more likely to win. On the other hand, sports and machine learning don’t have a strong relation, not yet. However, this is an obvious next step in sports evolution. The predictive models that come out of machine learning have tremendous power compared to commonly used statistical methods. 

In the past, I have worked directly with a team in MLS. The data was incredible and the results fascinating. There was data about the positions of players and the ball on the field measured in small (one to a few seconds) intervals. This allowed me to build a dataset that contained information about team formation, passing (long, short, forward, side to side, etc), running, and shots on goal. The model built was one that could model and suggest strategies that would increase the probability of shots on goal against specific opponents. This can be invaluable information for any team.

While the fluid nature of soccer matches makes this modeling, as well as what can be done with the results of the model, a bit difficult, football lends itself to machine learning beautifully. It is a game with many starts and stops, and at every restart, the teams have a new opportunity to make a new and independent play. Furthermore, the plays are defined and practiced. With machine learning, one could build models that would predict which plays your opponents would do in the next play. You could also have a model that can predict the probability of success for all your plays.

Football and Machine Learning

Let’s take a look at this Kaggle competition where you are asked to predict how many yards an NFL player will gain after receiving a handoff. There are a total of 49 different features, or columns, in the dataset for you to analyze. The features include very intuitive ones such as the current offense formation and not very intuitive ones like the jersey number of the player. Nonetheless, the machine might see some patterns that humans cannot observe and present us with a model that accurately predicts the yards rushed by a player. Imagine if the 49ers and Chiefs had such models in their position in their Super Bowl game. What would have happened? Maybe the 49ers could have stopped the last rush play by Damien Williams. Maybe the Chiefs knew that Damien Williams would run 38 yards just as the model predicted.

If you are an NFL team and would like to play in the Super Bowl LV, equip yourself with some machine learning power. Ople is here for you. We put the power of machine learning in the hands of analysts so you can make more data-driven decisions.