As always keep an eye on injuries and COVID Issues. All stats pull from the Last 10 games (L10), so be aware of teams stats that don’t reflect recent injuries, recent returns, and recent trades. I will regularly post updates throughout the day and annotate with a “*”.
*Residual absences from previous games could be skewing some results (for example, Zion, Lonzo, and Ingram missed a few games for the Pelicans within the L10 which could be negatively affecting their line projections since stats are pulled from L10).
Yesterday I got some good feedback, I will only post the top 3 scoring models, but I will keep track on how every model is doing and keep record up to date right here. I put up a feedback form at the bottom of the page, please provide some feedback if you want to see something different.
Sorry everybody you are just going to have to scroll through. I couldn’t use the clickable table of contents because I don’t have the premium version of wordpress, so its disabled (and I can’t afford to spend money just for a table of contents).
**Model Projections updated 9:40am CST to reflect line changes.
**I was notified that my O/U projection percentages for kNN were backwards, fixed and updated below at 10:15am.
Records
Here is how the models have been performing. So far most of the Machine Learning Models have performed well… but it’s been a small sample size.
ATS Record
O/U Record
Top Performing Models
Based on Average Win % between ATS and O/U.
- kNN – 75%
- Random Forest – 64%
- Ada Boost – 61%
- Neural Network – 55%
- SGD – 54%
- SVM – 52%
- Basic Statistical Model – 50%
- Linear Regression – 49%
Summary of Projections
ATS Projections
O/U Projections
Player and Team news
Teams on a Back to Back: 76ers (Philly to Detroit, Rough travel), Detroit (Home), Minnesota (Home), Miami (Toronto to Brooklyn), Brooklyn (Cleveland to Brooklyn), Chicago (Charlette to Chi), Houston (Detroit to Dallas, Brutal travel), Dallas (San Antonio), and Denver & Phoenix (played each other last night).
Kevin Durant was a late scratch from last nights game vs Cleveland, keep up on his status as game time approaches, although they could have been resting him for the back to back.
Jimmy Butler has now been ruled OUT! Tyler Herro is out. Both Heat and Nets on a back to back.
KAT is still out for the T-Wolves, but he has missed enough time for his impact to be mostly reflected in the latest projections.
Devin Booker appears to be a GTD to play for the Suns, That will be a huge impact if he is out. The line has actually moved to Den -2.5 since I first took the screen shots (indicating the books think he will be out).
Christian Wood is out for tonight and John Wall was upgraded to likely to play. John Wall’s return after missing a few games and Wood’s absence will not be accurately reflected in the stats.
The Pelicans, Heat, and 76ers stats may be skewed due to COVID issues keeping key players out of games within the L10 stat pull.
*Blake Griffin and Derrick Rose are ruled out for the Pistons.
*DeAngelo Russel has now been ruled out for the T-Wolves.
*Porzingas is looking like he will miss todays game for the Mavs, that wont be reflected in the stats.
*Booker is officially out for the Suns.
Sharp Report
Sharp money yesterday was 6-0 for ATS Picks and 1-5 on O/U’s.
Early Money
Not all of the bet stats are available yet this morning, but here is what I’m seeing.
ATS – Sharp money has come in on the Pelicans (-6.5), Warriors (+6.5, Small), and Nuggets (+1).
O/U – Sharp Money is on the Nuggets/Suns Under (219).
Mid Day
ATS – Sharp money on the Pelicans (-8) has slowed. Some small sharp money on Rockets (+10), Nuggets (-2.5) and Lakers (-9).
O/U – Sharp money has come in on the Pelicans/Timberwolves Under (221.5) and solid sharp money on the Nuggets/Suns Under (219).
Late Afternoon
ATS – Sharp money has steadied out on the Pelicans. Sharp money has continued on the Rockets (+9.5) with the John Wall news. Sharp money has also come in on the Pistons (+7.5).
O/U – Some Sharp money has come in on Lakers/Bulls Under (227).
Model Projections
1. k-Nearest Neighbor (kNN)
The lazy learner (That’s what my grad school prof called it). Nearest Neighbor has performed best with 30 Neighbors.
Record: 73% ATS, 78% O/U.
How it Works
[I will write this up later in the day if I have time].
Model Projections
The kNN model is off to a hot and fast start, let’s hope it keeps it going. The success has been too good for any model and this is likely do for a regression. Last season the kNN ended the year at about a ~61% ATS winning percentage.
Betting Value
If you don’t understand what you are looking at, please read my post about betting tips. The percentages represent value for that particular number. Good value alone doesn’t mean you should bet it (To quantify, >30% is great value, 20-30% is really good, 10-20% is decent, <10 is ok value, blanks are negative value).
2. Random Forest (R.F.)
[Nerd Talk] Best performance is capped at about 50 trees, anything more than that and the processing time/power required for a very minor improvement increases exponentially.
Record: 62% ATS, 67% O/U.
How it Works
[I will do a write up later in the day if I have time]
Model Projections
The Random Forest has been good at both ATS and O/U projections. Last year Random forest ended the season with a ~58% winning percentage
Betting Value
If you don’t understand what you are looking at, please read my post about betting tips. The percentages represent value for that particular bet type, number, and price. Good value alone doesn’t mean you should bet it (To quantify, >30% is great value, 20-30% is really good, 10-20% is decent, <10 is ok value, blanks are negative value).
3. Adaptive Boosting (Ada Boost)
I did not learn boosting in Grad School, but I know this is some form of boosted tree algorithm. But I had the code and the data I already had didn’t need to be re-formatted so we will see how this 100 tree estimator performs together.
Record: 77% ATS, 45% O/U.
How it Works
[I will do a write up later in the day if I have time]
Model Projections
The ATS predictions have been the driving force for this model and it has done very poor at correctly projecting Over/Under numbers. I did not use Ada Boost last year so we will see how it does in its first NBA season.
Betting Value
If you don’t understand what you are looking at, please read my post about betting tips. The percentages represent value for that particular bet type, number, and price. Good value alone doesn’t mean you should bet it (To quantify, >30% is great value, 20-30% is really good, 10-20% is decent, <10 is ok value, blanks are negative value).
why don’t you include player injuries/trades/Covid restrictions ? The model would be more accurate and since there’s so many injuries / covid restrictions the model is very inaccurate
Getting player data and putting it through a machine is fairly easy, although it would require some initial legwork. But going in and putting in who is in and who is out constantly, updating models and putting them up every time there is a change, and doing all of that for multiple sports is extremely difficult for one person to track and keep up with. I do this on the side and still have a full time day job (and family), there is only so much time I can dedicate to this. Maybe in the future if I can continue to grow big enough and make money from ads and other things than I can get somebody to help me and start going down that avenue along with DFS… but for now, it’s a bit too much for me alone to handle.