The Madness of March is upon us
Unfortunately, I may need to apologize (or possibly say you’re welcome if you aren’t even interested in the technical details of our models and analysis) because this is going to be a really short and sweet post. A deeper analysis will be coming covering the important games after all the crazy upsets happen. But I wanted to get this out quick tonight to show you which way A.I. Sports is leaning this year.
Monte Carlo Simulation
We simulated the entire tournament 100,001 times following a very similar approach we discussed with the NFL playoffs. Our code for the NCAA simulation can be found on our github. I told you this would be short and sweet so without anymore drama or build up…
The Results
It’s interesting that Duke has a higher probability of reaching the final game, but Gonzaga is more likely to win the entire tournament. This shows how important match ups are when trying to predict sporting events. Our models put Duke to win 2 out 3 games head to head against Gonzaga, but simulating the tournament as many times as we did helped us understand that Gonzaga has a better chance overall to walk away with the championship.
Our Approach
We have a multifaceted approach to this tournament. Obviously, our bread and butter is always geared towards the investing side trying to maximize our profits, but all of us are fans of the bracket. The bragging rights among co-workers, friends, families, and randos on the internet. Each one of us has our own bracket that includes a hit of modeling, a pinch of hometown bias, mixed in with a lot of our own personal opinion. We have also entered our algorithms into the Kaggle competition. This contest is hosted by Google Cloud and the NCAA geared towards data scientists squaring off algorithm to algorithm.
Last year our algorithm was sitting in 39th place out of a possible 1,868 unique algorithms after the first 2 weekends. I don’t want to get into it too much because it still hurts… but our algorithm may have needed Jesus because our agnostic mathematical approach left me cursing at a sweet, pure nun night after night.
Each Loyola win was like an evil sniper toying with me, slowly shooting out each knee, then arm, and with their final win sending them into the final four came the head shot to our algorithm and my pride. We finished 402nd overall. This year we will redeem ourselves.
Contest Rules:
We have to submit our algorithms predicting every single possible match up that could occur in the tournament before the first game is played. There is a total of 2,278 predictions with the associated predicted probabilities. What makes this contest so difficult is that you are not only judged on whether or not you correctly predict a win or loss. But you are penalized or rewarded based on the probabilities you assign to the game.
For example, our model last year had Loyola losing an early round game with an almost 95% probability. When they won, we were penalized heavily. On the other hand, Virginia was upset in the first round playing a 16 seed and while we predicted them to win, it was only at a 55% probability when others had them at 95+%. This played to our advantage.
The bracket that follows is our picks and associated probabilities for the contest. As the games move on I will update our bracket with the new probabilities from our contest submission. If you would like to follow along I will be posting it to our twitter account for simplicity. @theFirmAISports
Our Bracket:
I wish you all the best of luck with your own brackets and may the Madness of March bring joy to us all. Amen.
I think I still need to find some Jesus but I did promise myself not to curse out any nuns this year. 🙂
