2018 Men’s NCAA Basketball Tournament Follow Up

I know this is little late but better late than never, right? I was actually knee deep in my capstone and did not have time do my follow up until now. But here is how my bracket turned out:

I predicted Nova to win the whole tournament and Kansas to make it to the Final 4 but that is about the only positives. Both of those teams were one seeds, so that is not too impressive. One of the things I did want to look into my bracket about is why Gonzaga was predicted to lose in the first round to UNGC. When going back through my code I noticed that I had to drop duplicates one time when I was joining all my features together. I dropped them on “TeamID” and somehow another teams stats got attached to Gonzaga’s TeamID. I guess is that this happened to Houston as well. It is going to take a deeper dive into my code than I am able to do right now to find out exactly where it went wrong but I hope to be able to find out where before next years tournament.

Speaking of next years tournament I hope to take what I learned this year and applying to my plan of attack for next years tournament as well. I did a few things well this year and a few things I hope to change for next year. Some I of features I hope to use next year as well, are how long a coach has been coaching, rankings from Massey Rankings and conference rankings. I would like to use more rankings from a wider variety of systems next year though. I do have a lot more to change for next year rather than what I am going to keep from this year.

Next years plan of attack will be completely different. Some of the features I am going to use next year that I did not this year are the age of players on each team, narrowing down some of the box score stats to only use a few and conference tournament win/loss from the current year. I might want to create a feature called “Special Mascot/Person/Thing” to account for Sister Jean next year. The biggest change I plan to make it is how I set up information that my model will train on. This year I trained my model on games that happened in the regular season and next year I want to train my model in previous tournaments. An example if that I will train next years model on tournament games that happened in 2016 through 2018. I am hoping this will allow my model to pick up on more upsets and produce fewer vanilla predictions. I look forward to being able to try this for the years to come.

All of my code can be found at my GitHub at: https://github.com/CBJohnson30/NCAA-Basketball-2018