Zoltar is my NFL football prediction computer program. It uses a neural network and a form of reinforcement learning. Here are Zoltar’s predictions for week #10 of the 2023 season.

Zoltar:    panthers  by    0  dog =       bears    Vegas:       bears  by  3.5
Zoltar:    patriots  by    1  dog =       colts    Vegas:       colts  by    2
Zoltar:     bengals  by   10  dog =      texans    Vegas:     bengals  by  6.5
Zoltar: fortyniners  by    0  dog =     jaguars    Vegas: fortyniners  by    3
Zoltar:     vikings  by    6  dog =      saints    Vegas:      saints  by  2.5
Zoltar:    steelers  by    6  dog =     packers    Vegas:    steelers  by    3
Zoltar:      ravens  by    6  dog =      browns    Vegas:      ravens  by    6
Zoltar:  buccaneers  by    4  dog =      titans    Vegas:  buccaneers  by  0.5
Zoltar:     falcons  by    1  dog =   cardinals    Vegas:     falcons  by  0.5
Zoltar:    chargers  by    1  dog =       lions    Vegas:       lions  by    2
Zoltar:     cowboys  by    9  dog =      giants    Vegas:     cowboys  by 16.5
Zoltar:    seahawks  by    6  dog =  commanders    Vegas:    seahawks  by    6
Zoltar:        jets  by    0  dog =     raiders    Vegas:        jets  by    2
Zoltar:       bills  by   10  dog =     broncos    Vegas:       bills  by  7.5

Zoltar theoretically suggests betting when the Vegas line is “significantly” different from Zoltar’s prediction. In mid-season I typically use a threshold of 3 points difference but in early and late parts of the season I use a more conservative threshold of 4 points difference.

For week #10, using 3 points as the threshold, Zoltar likes three Vegas underdogs and two Vegas favorites:

panthers     at        bears : bet underdog panthers
texans       at      bengals : bet favorite bengals
saints       at      vikings : bet underdog vikings
titans       at   buccaneers : bet favorite buccaneers
giants       at      cowboys : bet underdog giants  OFF

The Cowboys (-16.5) vs. Giants game doesn’t count because Zoltar doesn’t handle huge point spreads greater than 14 points.

Using a conservative 4 points threshold:

saints       at      vikings : bet underdog vikings
giants       at      cowboys : bet underdog giants  OFF

Using an aggressive 2 points as threshold:

panthers     at        bears : bet underdog panthers
colts        at     patriots : bet underdog patriots
texans       at      bengals : bet favorite bengals
fortyniners  at      jaguars : bet underdog jaguars
saints       at      vikings : bet underdog vikings
packers      at     steelers : bet favorite steelers
titans       at   buccaneers : bet favorite buccaneers
lions        at     chargers : bet underdog chargers
giants       at      cowboys : bet underdog giants  OFF
broncos      at        bills : bet favorite bills

For example, a bet on the underdog Vikings against the Saints will pay off if the Vikings win by any score, or if the favored Saints win but by less than 2.5 points (i.e., 2 points or less). If a favored team wins by exactly the point spread, the wager is a push. This is why point spreads often have a 0.5 added — called “the hook” — to eliminate pushes.

This is the part of the season where injuries have a big effect on the point spread.

Theoretically, if you must bet $110 to win $100 (typical in Vegas) then you’ll make money if you predict at 53% accuracy or better. But realistically, you need to predict at 60% accuracy or better.

In week #9, against the Vegas point spread, Zoltar went 2-2 (using 3.0 points as the advice threshold).

For the season, against the spread, Zoltar is 28-20 (~58% accuracy).

Just for fun, I track how well Zoltar does when just trying to predict just which team will win a game. This isn’t useful except for parlay betting. In week #9, just predicting the winning team, Zoltar went only 8-6 which is terrible. Zoltar went 0-3 when predicting the winner of very close teams playing at home. Vegas was much better, going 12-2 at just predicting the winning team.

My prediction system is math-based in the sense that it computes a numeric rating for each team and then uses ratings to compute the predicted margin of victory of the better team. An entirely different approach for predicting NFL football scores is to use a sophisticated simulation program and then simulate a game many thousands of times.

Right: “Half-Time Football” by Lakeside was produced in the late 1970s and early 80s. It’s strictly a dice game and is medium complexity.