I Used Machine Learning to Assess QB Prospects the Last 3 Drafts

[Deets84]

I decided to test out my ML skills by creating a linear regression model designed to predict if a quarterback who had declared for the draft was going to be a franchise quarterback. Each year the results were calculated prior to the draft taking place. After three years of running the model, the results are posted below. Curious if anyone else here has experimented with using machine learning for prospect analysis before. 

 

2022			2023			2024
Kenny Pickett	37		CJ Stroud	60		Jayden Daniels	62
Carson Strong	34		Hendon Hooker	49		Bo Nix	50
Bailey Zappe	33		Aidan O'Connell	33		JJ McCarthy	50
Matt Corral	32		Bryce Young	33		Caleb Williams	40
Sam Howell	28		Will Levis	32		Drake Maye	33
Desmond Ridder	27		Dorian Thompson-Robinson	29		Michael Penix Jr.	31
Kaleb Eleby	21		Stenson Bennett	28		Joe Milton	31
Sam Hartman	21		Clayton Tune	27		Jason Bean	30
Anthony Brown	20		Jake Haener	27		Jordan Travis	27
Malik Willis	17		Jaren Hill	25		Spencer Rattler	27

 

Legend
50+	Projected Franchise QB
35 - 49	Possible Starter / Viable Backup
30 - 34	Possible Backup
< 30	Not NFL Calibur

 

Design Notes:

• Players cannot be rated until their college career has completed.

• When designing the dataset, there were some subjective elements in determining which quarterbacks would be used as examples of franchise quarterbacks. The mindset in selecting these quarterbacks was that they were players a team would not feel a need to upgrade from. So examples would be Allen and Mahomes. Players that were borderline or still early in the league were excluded (Like Tua and Lawrence). Most of the dataset included QBs that had been drafted but were clearly not franchise quarterbacks like Rosen or Winston. 

• I tested approximately 30 attributes and college statistics and selected 6 to include in the model that had a strong correlation with the QBs marked as franchise quarterbacks. The model then calculates scores based on how similar the QB prospect was to the franchise quarterbacks in those 6 key areas.

• I developed the numerical ratings based on benchmark players who were excluded from the model data. 

 

Result Notes:

• I compared the 2022 results to a player I considered the epitome of an average starter in Kyle Orton and the fact that all players in that draft were below him illustrated the point that it was a weak quarterback draft. Brock Purdy was not on the list of quarterbacks tested that year, however that omission compelled me to evaluate significantly more prospects in the subsequent years to avoid missing a key player in the future.

• The model strongly favored Stroud over Young in 2023, its correct selection there made me optimistic about its capabilities. Very curious to see how Hooker’s career turns out since he was just below the threshold.

• In 2024 it predicted that Daniels should have been the first overall pick over Williams. Williams’ score hints that he could have a comparable early career to someone like Trevor Lawrence. Bo Nix jumped out as an interesting datapoint to me since he was only a borderline first round pick when I utilized the model, but Denver took him close to where the results said he should be taken. 

[HTTRDynasty]

Am I supposed to read this as Joe Milton and Drake Maye have nearly the same chance of being successful NFL QBs?

Would have been great if you posted this before the draft. It may have saved our fanbase a lot of angst 😂

[NYRaider]

Over 60 basically guarantees that you're going to be goated. Hooker being that much higher than Caleb Williams and Drake Maye is crazy though

[Deets84]

8 hours ago, HTTRDynasty said:

Am I supposed to read this as Joe Milton and Drake Maye have nearly the same chance of being successful NFL QBs?

Would have been great if you posted this before the draft. It may have saved our fanbase a lot of angst 😂

 

I enjoyed getting Daniels & McLaurin at a bargain in fantasy 😆

 

And yep, the expectation is that Maye is not going to be a successful starter, hes going to get a lot more opportunities than Milton though. 

 

Hooker is definitely a big litmus test to see how accurate this is, the Lions appear high on him but no way he sees the field unless Goff gets hurt. Next year's draft also has a lot of potential first round QBs so very curious to see if the model has strong feelings about the choice between the top two QBs like it has the last two drafts.  

[minutemancl]

This is interesting. Please keep it up for a few more years so we can better determine how good your model is with more data.

Could you retroactively do this for QBs taken before you started running this, or is that how you determined the parameters of the model to begin with?

[MikeT14]

Early thoughts and questions:

Why is Sam Hartman in 2022?
At least your model knew 2022 was crap. I'd disagree with Howell, I think he's proven enough to be a backup. 
I like Nix, but even I am shocked at the 50. 

[jebrick]

Hooker has not shown enough to think he could be a possible starter.  The tricky part with him is the Tenn offense was so gimmicky you can't believe he has any of the skills needed by an NFL QB.