The Introduction
Analytics is a growing facet to understand the success of an unknown player. For running backs, we can take a data set of college statistics, athletic measurables, and draft capital to predict the success, and even the Fantasy Points-Per-Game that a college player entering the NFL.
Predictive Modeling is something that is new to myself. I learned a lot in several statistics, Computational Engineering Analysis, and Composite Modeling courses during my engineering collegiate duration. Today, I am a process engineer who began my fantasy football predictive analytics for sports gambling. I’ve chosen to take this into predicting rookie success. This is the first time I have built this out, but the results, on the surface, appear to show some of the best correlation that can be shown in predictive modeling.
We are using multiple composite modeling to utilize machine learning. When determining how to build all of this out, there are a set of questions that you need to take into account:
1) What do we want to measure?
2) What Data is applicable?
To answer these questions, it is up to the general population for what is important to measure in regards to fantasy football success. I, mostly, focus on the dynasty aspect of fantasy football. In regards to dynasty, it is important to look at the short-term payout of a player, and what the probability of doing that is. After playing around with several different ways to measure this out, I came to the idea of average PPR Points per game over the first three season’s of a player’s career. This, is a common way that is shown to predict player success, alongside odds of being a “Top X” player.
THE DATA
So to create the model we need to determine the inputs. What is important in actually determining success translation from college success, into NFL success? Well, the inputs are broken up into the following groups:
Draft Information: Overall Draft Pick
College Counting Statistics: Total Yards, Receptions Per Game
College Production Statistics: Self-Created Production Score (Weighted Yards/Touchdowns Per Age), Self-Created Efficiency Score, Total Dominator, Total Dominator Over Average
College Offensive Share: Total Yards per Offensive Attempt, MS Rushing Yards+Touchdowns, MS Yards+Touchdowns over Average, MS Receiving Yards
Athletic Metrics: BMI, Weight-Adjusted Speed Score
The Players Included
The model creation utilizes all players that are included from 2006-2018 NFL draft (257 individuals in total). Although the 2018 has not fully come into form, it is important to include their information as the utilization of running backs continue to expand in terms of usage accordingly. There was a time where players were eased in, but it seems as though utilization is happening at a quicker rate and normalcy is established after 2 seasons.
THE RESULTS
When inputting the information and utilizing several modeling techniques including linear regression, and others, we can correctly predict a players Fantasy PPG Seasons 1-3 within 3 points of actual at a 50% clip. Additionally, it’s shown that the player rankings of predicted PPG, matches this order of player rankings of actual PPG, at a clip of 82%.
This means that this can be an extremely useful tool in predicting NFL Players success as they come out of college and into the NFL.
THE ROOKIE 2020 CLASS
Taking this created model and utilizing towards the most recent upcoming draft, we find that Jonathan Taylor has the opportunity to be an Elite Fantasy Producer. The top-4 running backs on the model, are all considered top prospects by those evaluating these types of players though.
After that point, there are a few spicy takes that the model puts out. Having A.J. Dillon as the 5th ranked running back in the model over J.K. Dobbins something that I personally do not believe will be the actual outcome, but you can understand why with his analytics metrics appeal that shows in the model. Additionally, Joshua Kelley is a polarizing figure within the industry and those evaluating, but he shows up here as the seventh-ranked running back, even with his lower draft capital than others on the model.
SO, HOW DO WE USE THIS INFORMATION?
It’s important to know that the model will not predict everyone 100%, regardless of how close the correlation implies. There will always be outliers that are included, however, that is why they are outliers.
With this said, we can utilize this information to make intelligent decisions on players. If you believe that Josh Kelley can be one of the top running backs in the class, this model states that there is a chance that it could actually happen. If you are wanting Zack Moss to end up the top running back, the likelihood of this happening is very small, and you will most likely end up disappointed.
These do not reflect my personal rankings, however, it is the starting point of everything that I do. It’s important to know a players expected range of outcomes, and their probability compared to the others in the draft class when making these type of decisions.
I will continue to make adjustments, as this is my first year and opportunity doing this. I will continue to keep everyone informed and up to date as I continue to build this out.
