Here's a paper on predicting NFL QB success based on input variables and a machine learning algorithm. The most accurate prediction model that was derived is based on four inputs
1) collegiate win percentage,
2) college games started per season played,
3) body mass index (BMI), and
4) age at draft time.
The model derived turned out to be 88% accurate.
I tried data for Sam Darnold, Josh Rosen, Riley Ferguson, Kurt Benkert, Mike White, Luke Falk, Kyle Lauletta, Nic Shimonek and Logan Woodside. The only ones that came out as predicted success were Sam Darnold and Luke Falk.
The model used draft data from quarterbacks from 2000 to 2015 drafts. "One node of the decision tree found that quarterbacks with a win percentage less than 56.1% in his collegiate career had 96.6% chance of being a bust in the NFL."
The link has a computer graphic showing various quarterback's data. Something for the nerdy football fans.
http://duelingdata.blogspot.com/2017/04/predicting-qb-success-in-nfl.html - (
New Window )
Whats a bust? Whats a success?
Where did he publish the players used so we can see his numbers check out?
There wasnt alot of info given
I'm gonna guess that height may be the factor.
Regarding the paper cited in this thread, it is just some guy who submitted a paper (which was rejected). Basically any math grad student could do something like this. It can be done at different levels of sophistication.
In some applications we are just throwing every pieces of data we have at the machine and have the machine sort it out. In case we'd put in something like BMI just because we have it. (Note that 40 variables were put in and the machine selected 4, of which BMI was one of them).
In military problems I have seen substantial improvement in machine learning over a human attempting to figure out a model. If I were running an NFL franchise, I would definintely look into machine learning algorithms for player evaluation, although I'd want a serious professional to do it.
And it lists Teddy Bridgewater as a success?
Quote:
have to do with anything?
I'm gonna guess that height may be the factor.
Why not just use height? All these guys will have high BMIs. Unless you are extremely thin, your BMI is high. I did 6'4" 230 and that gets a BMI of 28.
Landon Collins at 6 ft 215 is 29.2.
6 ft 170 lbs is 23.1
Quote:
In comment 13873444 section125 said:
Quote:
have to do with anything?
I'm gonna guess that height may be the factor.
Why not just use height? All these guys will have high BMIs. Unless you are extremely thin, your BMI is high. I did 6'4" 230 and that gets a BMI of 28.
Landon Collins at 6 ft 215 is 29.2.
6 ft 170 lbs is 23.1
This is a machine learning algorithm, so the approach would be to put everything in (height, weight, BMI etc.) and let the computer sort it out.
Alex Smith was considered a bust for a long time. Mark Sanchez was considered a success for the first few years. Jets fans called him the Sanchize before the wheels fell off
This paper is a year old and he seems to be claiming Goff a failure. That assessment might be outdated now.
The algorithms sometimes find relationships between variables that we were completely unaware of.
wasn't he a bust in college, bouncing around perhaps screwing up programs (Colorado). There is much to be said about a "track record" and consistency. Did not see it in college.
No, but I'll bet that beyond the usual QB skills they will need enough mobility to do what Rodgers does to escape bodily harm.