DATA SCIENCE
2020 NFL Postseason Predictions from Machine Learning Model — Super Bowl
Chiefs are 1–2 point favorite, but model might be undervaluing Chiefs offense
Over the first three rounds of the NFL playoffs, I shared my random forest-based model (V 5.0) predictions (Wild Card, Divisional, Conference). While the model performed as expected after the first two rounds for predicting winners, it went 0 for 2 in the Conference round. Heading into Super Bowl LV, its cumulative performance is lower than anticipated(actual: 7/12, 58%; test set: 69.6%). However, when it comes to predicting score differences, across all three previous rounds, the model has outperformed other estimates shared by ESPN (supplied by Ceasars Sportsbook by William Hill; CSWH) and FiveThirtyEight (QB-adjusted Elo) [more details in Conference Review section]. For the Super Bowl, while all three predictions favor the Chiefs to beat the Buccaneers by 3 points or less, the results from the Conference round suggest that the Chiefs might actually have a bigger advantage than that [more details in Why V 5.0 might be undervaluing the Chiefs? section].
Below is the final set of predictions for the season from V 5.0, CSWH and FiveThirtyEight (that were available at time of writing):
Why V 5.0 might be undervaluing the Chiefs?
For the Conference round, I described how V 5.0’s prediction of a Bills victory was primarily driven by a difference in efficiency metrics. However, it turned out to be the biggest error by V 5.0 for the entire postseason (16.5 points). Could this just be due to typical variation in prediction error, or is it possible that V 5.0 has a systematic flaw? Upon closer inspection of the efficiency metrics [provided by numberFire] and taking into consideration the stellar performance of the Chiefs’ offense in the Conference round, I suspect the current Overall Offense value for KC might be an underestimate.
Both the Chiefs and the Buccaneers have been improving Overall Offense throughout the season, and going into Super Bowl LV, the Chiefs have the advantage. Though, consider that gains in this metric were modest between weeks 17–19 when KC rested many offensive starters, had a Bye week and had their MVP quarterback Patrick Mahomes and two of their top 3 starting running backs, Clyde Edwards-Helaire and Le’Veon Bell, miss significant time. In hindsight, the Overall Offense rating going into week 20 vs. the Bills seems like it was too low and might still be a little deflated due to lower output in previous weeks when key players were on the sidelines. Additionally, the defense did well against a potent Bills offense and they seem to be playing better in the playoffs as compared to the regular season.
Conference Review
CSWH and FiveThirtyEight pulled ahead in terms of predicting winners (8/12, 67%) and have locked in a co-victory according to that metric since all are choosing the Chiefs. When it comes to predicting score differences, V 5.0 remains ahead and has locked in victory for that metric given the predicted spreads.
Can the Buccaneers pull off the upset?
While all three models favor the Chiefs (and potentially are underrating them a little bit as described above), the Buccaneers have defied the predictions in the last two rounds as unanimous underdogs to the Packers and the Saints. In fact, according to CSWH and FiveThirtyEight, their chances for an upset are actually as good or better than they were in previous rounds. Their offense has improved after a slow start behind future Hall of Famer Tom Brady and their defense will be the toughest unit the Chiefs have encountered in the postseason. It’s worth noting that the Chiefs are dealing with injuries to the offensive line so they may have more difficulty than usual protecting Mahomes and creating running lanes for Edwards-Helaire.
Given the Chiefs offensive prowess, (barring a key injury) it’s tough to envision the Buccaneers winning by multiple touchdowns, but if a few things break their way, Tampa Bay may the first team celebrate a Super Bowl victory in their home stadium. Perhaps a timely turnover or an unexpected trick play could be the key to success.
Even if one team gets out to a hot start, either of these top flight offenses have the ability to score quickly, and mount a comeback. If the models are right, it will likely be a close game and may even come down to a game-winning drive.
