Popular Features

    The 2020 Machine Learning NFL RB Prospect Model

    The 2020 Machine Learning NFL RB Prospect Model
    Anthony Amico March 23, 2020 9:46AM EST

    What Can Machine Learning Tell Us About NFL RB Prospects?

    Last week, I released some machine learning model results that attempted to predict WR success. This week, I wanted to do the same for NFL RB Prospects. Evaluating an RB prospect before the NFL draft requires careful examination as draft capital is a huge part of their success. I studied 266 RBs drafted between 2004 and 2017. The goal, as always, was to predict how likely a back was to produce a 200 point PPR season within his first three years.

    The result was a two-model ensemble, using the following four components, listed in order of significance for an RB Prospect:

    1. ESPN Grade
    2. Adjusted Yards Per Play
    3. Binary Return Touchdown Check
    4. Speedscore

    The ESPN Grade was by far the most significant variable, making up 52% of the prediction value. Adjusted yards per play is yards per offensive team play, but with receiving yards counting twice. I tried something different with the return touchdowns versus what was done with WRs. Instead of crediting RBs more with more return touchdowns, I included just a binary yes/no value if they had a career return score. RBs with a return score hit over 2.5 times more frequently than RBs who do not.

    RB Prospect Model 2020 Results

    The results of the model for the 2020 RB class lie below:

    Jonathan TaylorWisconsin67.7%
    J.K. DobbinsOhio State55.8%
    Zack MossUtah40.0%
    AJ DillonBoston College37.6%
    D'Andre SwiftGeorgia36.8%
    Cam AkersFlorida State34.9%
    Clyde Edwards-HelaireLSU32.5%
    Antonio GibsonMemphis12.6%
    Eno BenjaminArizona State9.5%
    Anthony McFarlandMaryland5.7%
    DeeJay DallasMiami5.4%
    Ke'Shawn VaughnVanderbilt3.2%
    LaMical PerineFlorida3.2%
    Salvon AhmedWashington2.3%
    Levante BellamyWestern Michigan2.0%
    J.J. TaylorArizona2.0%
    Darrynton EvansAppalachian State1.5%
    Javon LeakeMaryland1.5%
    Raymond CalaisLouisiana1.3%
    JaMycal HastyBaylor1.0%
    Patrick Taylor Jr.Memphis0.9%
    Reggie CorbinIllinois0.5%
    Joshua KelleyUCLA0.5%
    Rico DowdleSouth Carolina0.5%
    Scottie PhillipsOle Miss0.4%
    Brian HerrienGeorgia0.4%
    Mike WarrenCincinnati0.2%
    Sewo OloniluaTCU0.2%
    Tony JonesNotre Dame0.2%
    Darius AndersonTCU0.1%

    We can be more confident in our RB projections at the top because early draft position and playing time typically yield success. It should surprise nobody that Taylor finds himself at the top here. He is perhaps the most prolific RB producer in college football history and will look to dominate in the NFL much like he did at Wisconsin. Taylor has a shot at a featured workload right away.

    Zack Moss may seem a little high here at three, but it is worth noting that his final season was almost as productive as Taylor’s. He ran just a 4.65 40-yard dash at the Combine, but caught 28 passes in his final season at Utah. He has three-down upside if a team takes him in the first 60 picks. It is worth noting that in private workouts, he reportedly ran a 4.58 40 yard dash time.

    After seeing Derrick Henry plow his way through the NFL’s final month plus playoffs, I’m sure teams will find themselves enamored with the size/speed threat of A.J. Dillon. Dillon played for a bad Boston College team but really carried the mail, with 300 or more carries in two of his three seasons. He also caught 13 passes last year.

    Perhaps my favorite RB prospect at the position is Antonio Gibson out of Memphis. He doesn’t rate particularly high in the model, but it is worth noting that he was mainly used as a WR in college, seeing just 33 carries in 2019. With that said, he averaged over 11 yards per tote, and scored four rushing touchdowns. After running a blistering 4.39 40 at 228 pounds, Gibson has a shot to be the David Johnson of this class.

    Popular Features