Introducing BALLER: A Predictive NFL Spread Model (Fantasy Football)
I am happy to present a new model to the Fantasy Footballer community that predicts the outcome of the spread of NFL games. This is a machine learning model that was trained on 13 years of historical data and tested on two. As every great model has a name, I am pleased to introduce everyone to BALLER*, which stands for Boosted Analysis and (aggressive) Learning (model for) Line Evaluations and Results.
This is in no way an official Fantasy Footballers betting model but hopefully a useful tool to make your decisions for fantasy and leans on certain spreads and who could cover.
Building the Model
This is an extreme gradient boosted model (xgBoost), which is a type of predictive model that intakes a number of different metrics in order to make each assumption in its decision-making process. While the math and programming behind this is a bit of a black box, the metrics that go into it are simple: for each game BALLER predicts, it notes both teams’ five-game rolling average for various statistics such as total yards, total yards allowed, average points for, average expected points, expected points per play, number of passes and rushes, etc. It also takes note of the Vegas lines for spread and point total going into the game. With all this input, BALLER can then make a prediction on what it thinks the spread should be put at; we’ll call this xSpread. We can then compare BALLER’s xSpread to Vegas’ spread to make a bet on each NFL game.
Evaluating the Model
While this is all fun and games, none of this matters if the model isn’t accurate! Luckily, it is. For those of you who have dabbled in sports betting, you might know that to beat the market, you have a success rate of at least 52.4%. Anything above that threshold is spectacular, as you will be raking in profits! Let’s take a look at BALLER’s accuracy on its testing dataset:
On predictions where BALLER expects a higher spread than Vegas (away team to cover), it has a 53.6% accuracy.
On predictions where BALLER expects a lower spread than Vegas (home team to cover), it has a 54.1% accuracy.
Stellar! Put the two together, and the model had an accuracy of 53.8%, beating the market by 1.4%!
In order to visualize these results, I rounded the predictions to the nearest five and plotted their expected spreads against the actual score difference following the game. Taking a look at how these points fall, BALLER’s predictions are incredibly accurate, as we see a linear relationship between the two variables.
Another way to analyze this model is to look at the percent of bets that were correct at each difference level between the xSpread and the Vegas line. We see that while differences closer to zero are less accurate, xSpreads which differ greatly from the Vegas lines tend to be more accurate.
Additionally, it is important to look at BALLER’s density plot–we see that the model has a very normal distribution of predictions, in fact, more normal than Vegas, which is a great sign, as it shows consistency.
Using the Model
And now, the moment we’ve all been waiting for…BALLER’s Week 1 picks! Below you will find the model’s picks for each game slated for Week 1. Keep in mind a couple of keys:
- It is using data from last season to predict these games — it will be better adjusted to each team following this week.
- It focuses on the spread of the home team.
- It is applied best on which team can cover the spread, not necessarily “use this spread” the model spits out.
Let’s break down some of BALLER’s best bets:
The Packers are in a great position to cover against Chicago in Week 1. Jordan Love has shown a ton of promise over the summer while the team has looked strong overall in camp. BALLER has them as 1.2-point underdogs, putting higher confidence in them than Vegas. Go Pack Go!
The Jets hype train is real. With Aaron Rodgers at the helm, anything is possible. As the Bills are known to start the season a tad shaky, BALLER’s prediction seems pretty desirable.
The Cowboys are heavy favorites in BALLER’s eyes. Putting them at 19.5 point favorites, the model is confident Dallas will run the Giants offseason momentum off the tracks and leave Daniel Jones’ squad in the dust.
That’s all for this article. Be sure to stay tuned for weekly picks from the BALLER model! If you have any questions about BALLER, feel free to reach out to me on Twitter!