Previously, we asked a neural network to predict which QBs and RBs would gain induction into the Hall of Fame. With the next class established, let's take a look beyond the Class of 2017. Which of those players already enshrined might not fit the bill? Who are the players who came closest without getting in? And which current QBs and RBs are on track or already a lock for the Hall?
Neural networks mimic how neurons work together in our own brains to answer questions and make predictions. (For more detail on the process, take a look at last week's predictions.)
Outliers in the Hall
Of the running backs and quarterbacks analyzed by our neural network, only a small handful are misclassified.
Most prominent is the model error on 2017 selectee Terrell Davis, whose election marks a distinct change in the standards voters have applied to running backs. Davis had only four full seasons of production before a knee injury derailed his career. Voters have previously valued longevity and yardage totals as primary criteria among RBs, and Davis' 7,607 yards weren't enough by those standards. But voters apparently placed a premium on his postseason success, a standard that had generally been applied only to QBs. Davis' Super Bowl MVP award and two rings were apparently enough in the minds of voters to make up for his shortened career. Davis' closest comparable is Gale Sayers, the legendary Bears RB whose star burned bright for five seasons before he suffered what was effectively a career-ending injury. But the model gets Sayers right despite his short span of production, largely because he was named first-team All-Pro in all five of his starting seasons, which sets him apart.
All of this isn't to say Davis should or should not be in the Hall, only that his election is without precedent. He's such a unique case, his inclusion as a teaching example to the model might not have much impact on future RBs, until we see someone else enough like him.
Another Bronco, Floyd Little, has Hall of Fame credentials that are notably lacking. Little ranks 59th among modern RBs in total yards (adjusted for era and season length), and he ranks 60th in total TDs (also adjusted) -- only 25 players ahead of him on those lists are in the Hall, and he is the lowest by far in either category among enshrined runners. The next-lowest member of the Hall in total yards is Larry Csonka, who was a frequent blocking-back and has over 30 percent more yards, and the next-lowest member in touchdowns is O.J. Simpson, with nearly 40 percent more TDs. (Incidentally, Little also has the lowest yards per carry of all inductees, although that was not ultimately found to be a decisive factor by the model.) Little's popularity might at least partially explain his election.
The other RB miss, according to the model, is John Henry Johnson, who retired in 1966 after 13 seasons with San Francisco, Detroit, Pittsburgh and Houston. He was selected for the Pro Bowl four times but never as an All-Pro. Johnson's yardage and TD totals are at the bottom end of the range for inductees, and he never led the league in either category. The only inductees with lower totals had five, three and two All-Pro selections.
The biggest outlier among QBs is George Blanda, whose career not only spanned 27 years but also three positions. He was a QB and a kicker and even started at linebacker in 1951. His QB chops alone might not have gained him induction (partially because his best years were in the early AFL, which is not fully credited by some voters), but the combination of his longevity and versatility must have made the difference to the voters who elected him in 1981. Blanda was such a unique player that his career only served to confuse the neural network.
One QB the model definitely gets wrong is Jim Kelly, a first-ballot Hall of Famer who falls outside the model for a couple of key reasons. First, the model doesn't know about Kelly's three years in the USFL. I also suspect that his four Super Bowl appearances without a victory are held against him by the model, when most reasonable people would probably see that for the accomplishment it is. In the model's defense, it isn't programmed to consider making the Super Bowl a good thing or bad thing. It just looks for patterns and connections to make the most accurate assessments. If I add just two years of Kelly's typical performance to his resume to compensate for his early career outside the NFL, the answer is a clear "in." Fran Tarkenton's Super Bowl disappointments are the most obvious comparison to Kelly's, but Tarkenton's career totals are far larger than Kelly's. Nevertheless, Kelly's criteria stays in the data as is, so the model will know how to handle similar careers.
Falling just short
Of the past candidates for the Hall who didn't (or haven't yet) been elected, a few stand out.
The running back candidates who come the closest but haven't made the final cut are Priest Holmes, Eddie George, Shaun Alexander and Don Perkins. Despite being a fantasy favorite, Holmes was the primary starter for his team in only five seasons, and despite three All-Pro selections he doesn't have the yardage or TD totals of most other inductees. George had sustained success over nine seasons but had few other distinctions and has only a single All-Pro year -- the tipping point for RB inductees seems to be two All-Pros on top of large statistical totals. The exact same can be said for Alexander. Perkins has one All-Pro selection and six Pro Bowl selections to his credit, but his statistical totals aren't Hall-worthy.
Among QBs, Ken Anderson is given a 17 percent chance of induction, which is a relatively large figure among those not elected, as the model typically produces highly confident answers near either 100 percent or 0 percent. Anderson's missing piece of the puzzle is a championship, and the other parts of his resume are not enough to make up for it. Earl Morrall, a 21-year veteran, two-time All-Pro and the victor in Super Bowl V with the Baltimore Colts, does not have the total yards or career wins to make the cut with an 11 percent chance.
Phil Simms is an intriguing case. The model gives him a confident "no," but had Simms not been injured and won the Super Bowl in the 1990 season rather than his backup, Jeff Hostetler, he would very likely be in the Hall by now. Adding an additional Super Bowl appearance and win to Simms' record gives him a 97 percent probability of induction.
The other notable QBs who were given some chance of induction are Rich Gannon, Jim Plunkett, Daryle Lamonica and Boomer Esiason.
Future RB candidates
Edgerrin James was a semifinalist this year, but his chances of eventual election are poor. James' longevity, yardage and TD totals are comparable to many other inductees, but he has just one All-Pro and three Pro Bowl years.
LeSean McCoy has quietly amassed a potentially Hall-worthy record, at least by historical standards. His rushing and scoring totals are relatively low for an inductee, but they compare well at the lower end of the range. With two All-Pros and five Pro Bowls in eight years, the model would favor him strongly with a couple of more years of production. He just finished one of the best seasons of his career, and if he stays healthy, his career totals and awards will make it difficult for voters to leave him out.
Frank Gore has an excellent Hall portfolio as well, which the model assesses at 82 percent if he were to retire today. Gore has the longevity plus the yardage and touchdown totals to make up for the fact he never led the league in yards or TDs and was never selected as an All-Pro.
Adrian Peterson's career is indisputably Hall-worthy by almost any measure, and the neural network doesn't disagree, giving him the highest probability of any RB, either in or out of the Hall, at nearly 100 percent. He has a chance to move into the top five in total yards and total TDs (each stat adjusted for era) in 2017 -- that and Peterson's four All-Pro selections seal the deal.
Future QB candidates
What about the top QBs currently playing or recently retired? For starters, I cheated a bit and assumed Peyton Manning's and Tom Brady's careers are certainly Hall-worthy. Since the set of inductees is relatively small, every extra bit of information helps when teaching the neural network. Beyond them, the model thinks Drew Brees, Aaron Rodgers and Ben Roethlisberger are virtual locks for the Hall, even if they never take another snap.
Eli Manning is a more debatable case, with a 72 percent chance of induction. Beyond his two rings (including MVPs in both Super Bowls), he has impressive career totals and was one of the league's best passers for a four-season stretch, with Pro Bowl selections each of those four seasons. His interception rates are relatively high for his era, but stats like that haven't had much weight in the collective judgment of the voters, especially years after a career has ended and its lowlights slowly fade from memory. However convincing Eli's case might be in isolation, it might ultimately pale in comparison to his contemporaries, and a logjam of legendary quarterbacks could keep him out. The model only measures how well a career compares to others' in the Hall and isn't able to account for a contemporaneous glut of superstars, at least not yet.
Of course, another Super Bowl win would give Eli a 97.7 percent chance of induction.
Other notable veterans need to do more to break into conversation for the Hall. Philip Rivers is a top passer with enough yards, wins and awards, but he is missing a ring. If Rivers either wins a ring or adds another All-Pro caliber season, the model thinks he'd be a likely inductee. Tony Romo is missing a ring and enough top awards. One All-Pro season would put him in the conversation with about a 60 percent chance, but he would also need a ring to clinch. Carson Palmer has the longevity and the passing yards, but not much else beyond his three Pro Bowls.
Joe Flacco is on the other end of the spectrum. He has the longevity and postseason success needed, including a Super Bowl MVP, but lacks any season-level awards to be a credible candidate.
Taken just 15 picks before Flacco in the 2008 draft was Matt Ryan, who has all the career credentials except winning a Super Bowl -- which he very nearly did Sunday. Had he won in Houston, Ryan would have gone from a near-zero chance to a near-100 percent chance at the Hall. It might not have felt that way in the moment, but five years after his career is over, perhaps 10 or more years from today, the importance of a Super Bowl win might seem different to voters. Ryan has the yards, the wins, the All-Pro selection and four Pro Bowl citations, and he would then have enough playoff success to put him in a league with most other passers in the Hall.
Donovan McNabb has a credible case. He has the longevity, the total yards, the wins, a Super Bowl appearance and even six Pro Bowls to his name. But his overall record falls short with a 36 percent shot -- substantial support, but unlikely to garner enough votes. Like many other near-greats, one All-Pro year or a ring would have put McNabb over the top. You might argue with him as a credible candidate (as I would), because his individual skills didn't impress nearly as much as his contemporaries, but our argument isn't with the neural network. It's with how the Hall voters, as a group, have voted in the past. McNabb's ugly end to his career in Washington and Minnesota might still be fresh in our minds, but it might seem like an afterthought several years from now.
A few in the newest generation of top QBs might be considered to be on track. Cam Newton led his team to the Super Bowl with an All-Pro 2015 season, but he would need to basically repeat the success of his first six seasons to put himself in -- something that seems very possible. Russell Wilson is another QB we can say is on track for the Hall . He would need to continue playing at a high level and either win a second ring or notch an All-Pro year. Andrew Luck would need to do even more. His three Pro Bowls are the only thing going for him so far with respect to Hall of Fame criteria. Luck would need to continue playing at a level near that of his first three seasons and win two rings, two All-Pro selections, or one of each.
For more from ESPN Analytics, visit the ESPN Analytics Index.