Getting Started
How are these predictions made?
I use a combination of adjusted efficiency metrics, player based stats, some feature selection/augmentation, and then feed these in to a few different machine learning models. The numbers you see are the outputs of my highest performing models during backtesting.About the rankings
First off I need to credit Bart Torvik and his ranking site for allowing me to put something like this together. It can be viewed here: Torvik RankingsUsing some of his explanations in the FAQ section on his site. I was able to create a ranking system similar to his and Kenpom's. I don't believe my rankings are better than Kenpom but since the adjusted effeciency metrics play such a key role in the final model I was willing to sacrifice some quality to have full control over how they were being made.
Adjusted efficiency metrics are a stat used to estimate how many points a team would score (offense) or allow (defense) for every 100 possessions against an average D1 team. The adjEM or adjusted efficiency margin is just the adjusted offensive efficiency - adjusted defensive efficiency.
What is projected spread and what should I bet?
Projected spread is the spread that my model is predicting.For example: Illinois -15.3 means my model has illinois winning by 15.3.
Iowa +12.1 would mean I have Iowa losing by 12.1
Bet when you see the fire 🔥 emoji.
Lines are constantly changing. Right now I have a difference of 4 for spreads and a difference of 2 for totals as the cutoff point. For totals I believe there is a strong correlation between distance from the betting line and edge. I would say bets more than 4 points away are stronger and 6 or more are strongest.
If you want to follow this as intended, make sure the bet still falls within that threshold. I would stay away from betting games right before the start. The lines are extremely efficient at close and I don't believe the model can beat the closing number without a stronger bet / no bet criteria.
Why are some predictions so far away from the lines?
Typically these bets are all over the place at the beginning of the season. As the season goes on and I get more data on the teams the predictions will improve. For now I would proceed with caution.What is better, spreads or totals?
Totals are definitely better.How good actually are these predictions?
I did a complete rework of the model this year so last years performance won't tell you much. I will eventually have another page on this site to view a breakdown of how the model is doing this year. For now you will have to track for yourself.How much should I bet on any given pick?
Short answer: I am not sure. If you can bet the open, bet more than usual. If the line hasn't moved against me by the close I would stay away or bet smaller.Long answer: I will probably devote some time at some point later in the season to optimizing actual numbers. I typically bet 1% or less of the bankroll.