The problem sometimes with explaining machine learning outcomes to NFL general managers, scouts and coaches, regarding NFL quarterbacks, is that the results are black boxed and it is hard to understand why the computer selects certain players over others. This article is going to try and put some visualization to players to demonstrate how some great quarterbacks compare to some less than great quarterbacks based on college statistics.
It is pretty easy to come up with a list of great quarterbacks that have played in the NFL, but it is hard to compare the college stats of older generations in today’s more aerial based game. The list below will be comprised of some of the more prolific passers with successful NFL stats and some of the projected quarterback draft picks who did not pan out in the NFL.
By looking into these stats, we hope this helps you see what our NFL Quarterback Machine Learning System sees when it analyzes players’ college stats. The machine learning clustering will relate the quarterbacks in the clusters based solely on college statistics.
Not So Good
Here is the graphic representation of the college statistics and there relation to all the players in both lists.
click image to see more detail
Here is the textual version of the graph showing the 5 clusters.
Cluster 1: manning|peyton, carr|david, leinart|matt, brees|drew, roethlisberger|ben, manning|eli, rivers|phillip, luck|andrew, prescott|dak
Cluster 2: brady|tom, aikman|troy, young|vince, rodgers|arron, sanchez|mark
Cluster 3: marino|dan
Cluster 4: ponder|christian, dilfer|trent, harringotn|joey, russell|jamarcus
Cluster 5: quinn|brady
We think that this provides great insight regarding how college statistics can predict NFL success. Cluster 1 is particularly interesting how the names (except Carr) are very elite. Yet cluster 4 and 5 are not. Cluster 2 shows some very elite quarterbacks as well.