Predictive Modeling and Analysis for 2020 Wide Receivers
First I will describe my model. Skip ahead if you just want to see the results
The first thing I wanted to attempt while evaluating the 2020 wide receiver class is to create a predictive model with as many variables as I could find. I started by looking into SPARQ scores on the website Three Sigma Athlete. I used the data from this site for the wide receivers from the years 2014-2017. I removed all the players that are not currently in the NFL, and for each player, I added the stats for the year in college where they had the most yards, as well as their college dominator score found on the website player profiler. I did the same for the receivers in the 2020 class.
I had to train the data set to somehow predict how good a player will be in the NFL, so I had to somehow assign a rating to the current receivers in the training set. To do this, I input the current madden rating (lol) for each player and equally weighted it with my own personal rating that I gave for each player. This resulted in a score that I scaled-down to range from 1-10.
Next, I had to decide the exact variables I wanted to use in the model. The variables I ended up using to train the model were, in no particular order: height, weight, arm length, hand size, pSPARQ (as found on three sigma athlete), 40 time, vertical, broad, dominator score (found on player profiler), total yards in the best year of college, and yards per target in this same year. I definitely wished I was able to use agility tests, such as shuttle or 3 cone, but many of the 2020 prospects did not have times for this. If there are any more possible inputs or a different combination that you think would be better, let me know. I did test around with different combinations and manipulations of the input variables but decided to use all of my variables straight up.
I also had to decide on what machine learning algorithm to use. I messed around on this for a while, and in the end, I thought a K-nearest neighbors test is the most logical way to go about it. If you are not familiar, this test basically takes a test observation and gives it the average value of the K number of nearest neighbors. In my model, I had a K value of 3, so the 2020 receiver would be graded on the average result of its 3 closest observations. I thought 3 was a fair number to use for my K value, as my model itself gave it as an optimal value and I think it is reasonable seeing my small training set of 48 wide receivers.
Here is the code for reference and my results. Note that a couple notable prospects were not included in this model because I could not find sufficient data for some of their combine tests.
Model Interpretation
The first thing I realized is that this model is not predicting which players will produce at a high level in the NFL; rather it is telling me which players have the numbers and stats that are very similar to NFL players who have either succeeded or not succeed in the league. For example, my model is telling me that Brandon Aiyuk has the profile, combine numbers, and college production that is very similar to elite NFL players. On the other hand, Tyrie Cleveland has the profile, combine numbers, and college production that is similar to lower level NFL receivers. It may be slightly suggesting it, but my model is not straight up saying, “Brandon Aiyuk is the best wide receiver and is clearly better than CeeDee Lamb”. To give you examples of scores I assigned players in the training set, Michael Thomas and Deandre Hopkins scored the highest at 9.95 and 9.90 out of 10, while Eli Rogers and Alex Erickson scored lowest at 5.35 and 5.375.
Further Statistical Ranking
An additional area of interest for me was further looking into the college production of each receiver as opposed to heavily weighing their athleticism and build. By using the great cfbscrapR package, I found the number of Big Plays for each player, defined as an offensive play that gained over 25 yards. I also found the number of Big Plays per reception, to put each player's opportunity on a more level playing field. This sort of acts as an explosiveness rating for each player. In general, one can assume better players make Big Plays more often than weaker players. QB play, coaching, and luck can be a factor in Big Plays of course, but Big Plays per reception removes some of the noise.
Next, I found the total EPA for each player, which is the expected points added that each player generated on offensive plays over the course of the season. Please research this if you want a better explanation, I am not entirely sure how this is calculated, but it seems legit. Lastly, I also calculated the EPA added per reception for each player. The EPA per reception value could be skewed by Big Plays and touchdowns, but the overall value can help the underneath/middle of the field grinder type receivers that are less explosive. I was unable to determine drops, so these were not included in the EPA numbers.
Here are the full results for Big Plays and EPA:
Player Notes, Model Outliers, and Statistical Standouts
Brandon Aiyuk
· 6’0”, 205 lbs, Sr., Arizona State
· 1st in model
· 6th with 15 Big Plays (3 70+ yd TDs)
· 16th in total EPA
Aiyuk was the number 1 ranked player in my model by a decent margin. This was not particularly expected, but its fair to say that its not hard to see his overall profile as one that has had success in the NFL. He had a lot of Big Plays, and out of these big plays, many were Huge Plays. The explosion seen in these 50+ and even 80+ yard touchdowns is certainly exciting. I can’t find anything bad to say about Brandon Aiyuk, I’m high on him just like my model. So, I guess we’ll just have to wait and see if the model is right.
Justin Jefferson
· 6’1”, 202 lbs, Jr., LSU
· 20th in model
· 11th with 14 Big Plays
· 2nd in total EPA at 73.3
Among the consensus top 10 or so wideouts, Jefferson was one of maybe two or three of them that my model ranked noticeably low. Did my model sniff out that Justin Jefferson plays for the best QB, and best college team, and that his perceived skills and numbers are inflated? Yea no, I doubt it. That’s just my only possible explanation for his low ranking in my model. He had a decent amount of big plays, but nothing too standout. Given the insane skill and improvisational skills of Joe Burrow, I’d probably have expected more Big Plays from Jefferson. Jefferson was a standout in total EPA, ranking second with 73.3 total EPA. This does play to his profile as a high catch rate, middle of the field type receiver, so that makes sense.
Jerry Jeudy
· 6’1”, 193 lbs, Jr., Alabama
· 8th in model
· 16th with 10 Big Plays
· 7th in total EPA
In discussion for the best receiver in the draft, it is slightly surprising that Jeudy didn’t light up the scoreboard in either my model or rankings. But again, the model’s purpose is not to single-handedly predict who will be the best, rather it shows who’s projected for success based on the success of receivers with the most similar profiles. And there is a clear reason why Jeudy may have produced a moderate score in both the model and all the rankings: his insane quickness and route running. Quickness and agility were unfortunately unable to be accounted for in my model. Additionally, a skill set involving the best route running and insane quickness is especially good for shorter or moderate routes as opposed to long balls, and even more true when your opposing wideout is an absolute burner and can lengthen the defense for you. So, there is some reason why Jeudy may have produced relatively moderate results here. In my opinion, Jeudy is the best receiver in the draft.
Henry Ruggs III
· 5’11”, 188 lbs, Jr., Alabama
· 3rd in model
· 14th with 11 Big Plays (3 70+ yd TDs)
· 2nd in EPA per reception
I couldn’t not write about Henry Ruggs, he is a beast. Absolute burner looking to be the next Tyreek Hill or Mecole Hardman. An overall elite athlete as he is leading the class in SPARQ. His speed shows up with three 70+ yards touchdowns and ranking 6th overall in Big Plays per reception. Also notable is that he is 2nd among 2020 receivers in EPA per reception. What does that mean? Get the ball in Ruggs’ hands and good things will happen. This EPA per reception could be skewed by touchdowns or something, but all I know is Ruggs is an explosive player and his high ranks in both per reception stats means that when he gets the opportunity he’s probably going to make a big play.
CeeDee Lamb
· 6’2”, 198 lbs, Jr., Oklahoma
· 11th in model
· 1st with 21 Big Plays
· 9th in total EPA
CeeDee is obviously a monster. He blows away the field in Big Plays and Big Plays per reception and some of his plays are eye-opening. I was slightly surprised at his moderate ranking in my model given his overall balance of size, speed, and power. There is not much else to say other than I wish GM's luck at picking between CeeDee, Jerry Jeudy, and Henry Ruggs.
Isaiah Hodgins
· 6’4”, 209 lbs, Jr., Oregon St.
· 2nd in model
· 21st in Big Plays
· 5th in total EPA
Being ranked 2nd in my model, I wanted to talk about Hodgins. He is a bigger, lankier receiver which means he didn’t show up well in Big Plays, but he ranked high in total EPA. Obviously, he’s not one of the top receivers in his class, but he stood out in 2 of my rankings. One thing I noticed when watching him is he has great hands and a knack for reeling in the ball. If he could improve his quickness and route running, he could be good. Because of all this, I’m fine with Hodgins in the mid to later rounds relative to peers among the second or third tiers of receivers in this class. Again, we’ll see if the model indicated something with the high ranking of Hodgins (probably not).
Chase Claypool
· 6’4”, 238 lbs, Sr., Notre Dame
· 16th in model
· 12th in Big Plays
· 11th in total EPA
I wanted to write a note on Claypool because in some of my initial models he kept ranking really high, 1st in multiple of them, and that caught my attention. In the final model though, which I think is best, he ranked 16th. He’s a big target and has respectable speed; I don’t like comparisons, but he reminds me of Mike Evans. He doesn’t quite top the charts in Big Plays or EPA. I like Claypool.
Laviska Shenault Jr. and K.J. Hamler
These are two prospects that I have seen some people rank fairly high. Unfortunately, I was unable to get them in the model because I don’t think they had numbers for some combine drills, etc. Both ranked moderate or even low in Big Plays and EPA. Laviska is big and strong. It will be interesting to see how a tank like him translates to the NFL. Hamler is really fast. Kind of like a discount Henry Ruggs. I like Hamler a little more than Laviska.
My Personal Final “Rankings”
As an NFL GM with an average offense with no specific strength, weakness, or need at wide receiver, here is how I would rank the 2020 class after my modeling, analysis, and highlight-reel watching (I watch very little college football). Jeudy, Ruggs, and CeeDee are clearly the top 3 in my opinion, and very hard to rank. Choosing between them largely depends on what the offense needs.
1) Jerry Jeudy
2) Henry Ruggs
3) CeeDee Lamb
4) Brandon Aiyuk
5) Denzel Mims
6) Tee Higgins
7) Chase Claypool
8) Justin Jefferson
9) K.J. Hamler
10) Jalen Reagor
Let me know if you have any suggestions for my model or analysis, or would like to see my data. Thanks for reading my “expert” analysis.
