Bracketology: Picking Rounds 1 & 2 Using KenPom Stats! 

Hi Vince! Thanks and welcome! Looking forward to seeing the selections on Sunday. Cheers!

Hi Jaxdog! Thanks for your support. I will be posting a similar video (to this one) detailing how to use decision boundaries with KenPom stats to predict the final four and the overall champion. Stay tuned! Cheers!

Hi King! I'm glad you like the video. I agree. I just posted a video on how to find upsets. It uses randomization of KenPom probabilities to pick a few upsets. We'll see how it works. Cheers!

Hi 2DISCiples! Sure thing. The blue slanted line is AdjO = AdjD + 22.9. So for teams to be above this line, that implies that AdjO - AdjD > 22.9, or simply AdjEM > 22.9. The orange line is AdjO = AdjD + 10.0. Hope this helps!

Hi Minhajghayur! The KenPom stats that I use and discuss in my videos are all *pre-tourney* stats. Cheers!

@@KerrySportsAnalyst Thank you for confirming! This content you’re creating is great! looking forward to Tomorrow’s video

Hi Zachary! I totally understand the confusion here. These are conflicting stats. When this happens, I resort to the matchup. According to the KenPom prediction formula, Kentucky has an 86.2% chance of defeating Oakland. So I am picking Kentucky. Hope this helps!

Hi Adam! There are currently 13 teams with AdjEM > 22.9 as of 9pm on Thursday night. I definitely think the conference tournaments will give us more teams that satisfy these boundaries. Also, when the final selections are made, we'll see more teams below the lower bounds. However, you are absolutely correct. There will be games with two teams in the "middle" of the boundaries. For those, I plan on using the following KenPom spreadsheet from another RU-vidr. It is free to download. I might also use my own KenPom model I created with Python. I just haven't been thrilled with the results, yet. Cheers! Here is a link to the Excel KenPom model video for single game predictions: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-N0fvKpe2sJQ.html

Hi Tyler! I created that chart using the pre-tourney KenPom stats from 2001 - 2023. I use Python and its various libraries to do almost all of my analysis. I would be happy to share the code if you are interested. Cheers!

@@KerrySportsAnalystI would actually be interested in looking at the Python code if you don’t mind sharing

Hi CD2_! That is a very good question. My video does not address how to pick games "inside" the decision boundaries. Part of the answer will be in my next video where I use decision boundaries to pick the final four and the overall champion. However, these teams are likely to be above the high values. For the "in-between" teams I plan on using a KenPom single game prediction model created by another RU-vidr. It is an Excel spreadsheet that is a free download. My plan is to be "all-in" on KenPom this year. I actually believe the strength of this strategy will be in predicting the overall champion. We'll see how it plays out! Cheers! Here is a link to the Excel KenPom model video for single game predictions: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-N0fvKpe2sJQ.html

Hey Levi! That's an interesting idea. I've trained machine learning models to predict wins. But the variance in March Madness is hard to predict. Cheers!

Is there any concern of overfitting by using the entire available data to produce the decision boundaries? Also, maybe it's more worthwhile to use say the 5 most recent years to pick up on any new trends.

@@mikebabiak Hey Mike! That's a very good question. Overfitting is always a concern. However, after training a model, it is common practice to train the "final" model on all available data before using it to predict future events. That's how I view this. I'm trying to predict 2024. If you ignore where the decision boundaries came from for a moment, here is how the decision condition "AdjEM > 22.9" played out from 2018 - 2023 in the first round: 2018 - 90.0% (9 of 10), 2019 - 92.3% (12 of 13), 2021 - 88.9% (8 of 9), 2022 - 84.6% (11 of 13), 2023 - 88.9% (8 of 9). There are currently (3/14/2024) 12 teams with an AdjEM value greater than 22.9. Cheers!

you have to pay for it

Hi MaverickFan! While I do not necessarily recommend paying for data, the annual subscription fee for KenPom is aroud $22. Also, you can see the pre-tourney data for 2024 before the games start today (3/21/2024). Cheers!

Hi Kevin! I agree. Especially when it comes to predicting the higher seeds. I think the upsets are exceptions to that rule. But it would be interesting to see if a team with a lower AdjEM would still be predicted to win if they have, say, an elite AdjD and their opponent has a mid-level AdjO. I'll definitely look into this. Cheers!

Hi Rasak! That surprised me too. Their AdjD for 2023 was 118.425 which was third worst among all D1 schools. Yet the pulled the upset. Wild!