Looking at Wide Receivers Analytically: 2018 Edition

Last year, I made a post on this blog called Looking at Wide Receivers Analytically, which used regression analysis to project college wide receivers into the NFL. I had intended to keep this series going with other positions, but things came up that prevented me from posting my findings. This year, I have decided to update that model and also use another machine learning algorithm to evaluate this year's prospects. For the sake of brevity, if you want to see the linear regression model explained, take a look at last year's post. That model is almost the same, with only some slight tweaks being made. 

The new algorithm that I will be using is called a "random forest" model. In short, a random forest model uses many decisions trees in order to make predictions based on data. Think of a decision tree as a flowchart, sorting every prospect into different groups based on their production and workout data. This is what a single tree in this "random forest" can look like:

All of these decision trees are used on each prospect, and the average of all the trees in the forest is what is used as a prediction for each prospect's NFL career. As with the linear regression model, we are using approximate value per season as the production metric we are looking to find. AV isn't a perfect metric by any means, but I would like to be able to compare across positions once this project is finished and this seemed like the best way to do it. The parameters used in the random forest are market share of receptions, yards per team attempt, TDs per team attempt, career rushing yards, career kick return yards, career punt return yards, a logarithmic transformation of draft position, "breakout" age (age when a prospects records >35% of team yards), weight, broad jump, and 3 cone time. 

In order to rank this year's prospects, I took the average of the linear and random forest models. The draft position is a very rough assumed number (CBS's .... interesting prospect list this year makes it harder to get an actual number for each guy). Here's the top 10 list:

Name	Linear Model	Random Forest	Average
DJ Moore	6.35	5.24	5.80
Anthony Miller	3.54	5.18	4.36
Christian Kirk	4.56	3.94	4.25
Courtland Sutton	4.36	3.28	3.82
DJ Chark	2.87	4.03	3.45
James Washington	3.07	3.39	3.23
Calvin Ridley	2.50	3.28	2.89
Michael Gallup	2.38	3.15	2.76
Tre'Quan Smith	2.16	3.25	2.70
Daurice Fountain	2.27	2.38	2.33

Since 2014, there have only been 5 wide receivers with a higher grade than DJ Moore: Amari Cooper, Corey Davis, Corey Coleman, Sammy Watkins, and Odell Beckham. Calvin Ridley's number would the lowest for a 1st round receiver since 2014. The next 5 lowest are Cordarelle Patterson, Phillip Dorsett, Mike Williams, Breshad Perriman, and Laquon Treadwell.

As for late round gems, Tre'Quan Smith, and Daurice Fountain seem to top the list. They place 10th and 12th respectively out of 72 ranked prospects taken in the 4th round or later since 2014. That list includes guys like Dede Westbrook(1st), Jamison Crowder (4th), Tyreek Hill (7th), and Stefon Diggs (9th).