Announcing: SportsRatings' new Genetic Model for NCAA prediction
We've been working on a new model for NCAA prediction, and the preliminary results are in. The model tests several dozen different "genes" (i.e. heuristics) that relate to NCAA basketball, from power ratings to polls to combinations of different data, combines them together as "offspring", then kills off the weaker ones so the stronger genes in the "population" survive.
In the end, the model outputs a series of heuristics and associated weights on which to score the teams in the field. For the 2018 season, the initial results are below, with the "score" shows how close the teams are in the model's estimation. After running the results through the 2018 bracket, the "Finish" column shows how far the teams are expected to advance.
According to our model, we're in for a repeat of the 2016 final, with Villanova beating North Carolina. Duke makes the Final Four and the Arizona Wildcats crash it, too. Villanova's score is pretty far ahead of the Tar Heels, but from there it's pretty tight.
A few teams rank very high but the bracket isn't their friend. Top-ranked #1 seed Virginia is 4th in the model, but meet Arizona in the Sweet Sixteen and is eliminated; Another key Sweet Sixteen matchup is Duke vs. Michigan State; should the Spartans win they'd be favored to make the Final Four. Top seeds Kansas and Xavier only make the Elite Eight.
| Seed | Team | Genetic Rank | Score | Finish |
| 1 | Villanova | 1 | 99.4 | Winner |
| 2 | North Carolina | 2 | 74.0 | Runner-up |
| 4 | Arizona | 3 | 69.6 | Final Four |
| 1 | Virginia | 4 | 64.9 | Sweet 16 |
| 2 | Duke | 5 | 62.8 | Final Four |
| 3 | Michigan St. | 6 | 61.9 | Sweet 16 |
| 1 | Kansas | 7 | 59.4 | Elite Eight |
| 2 | Cincinnati | 8 | 56.9 | Elite Eight |
| 1 | Xavier | 9 | 56.7 | Elite Eight |
| 2 | Purdue | 10 | 54.2 | Elite Eight |
| 4 | Gonzaga | 11 | 48.0 | Sweet 16 |
| 3 | Michigan | 12 | 39.1 | Sweet 16 |
| 4 | Wichita St. | 13 | 38.2 | Sweet 16 |
| 4 | Auburn | 14 | 33.4 | Sweet 16 |
| 3 | Tennessee | 15 | 32.2 | Sweet 16 |
| 6 | Florida | 16 | 31.5 | Sweet 16 |
| 6 | Miami FL | 17 | 31.2 | Second Round |
| 3 | Texas Tech | 18 | 31.2 | Second Round |
| 5 | Kentucky | 19 | 29.9 | Second Round |
| 5 | Clemson | 20 | 28.7 | Second Round |
| 6 | TCU | 21 | 28.2 | Second Round |
| 5 | West Virginia | 22 | 27.5 | Second Round |
| 8 | Seton Hall | 23 | 25.0 | Second Round |
| 5 | Ohio St. | 24 | 23.7 | Second Round |
| 6 | Houston | 25 | 23.0 | Second Round |
| 7 | Nevada | 26 | 18.0 | Second Round |
| 7 | Texas A&M | 27 | 17.8 | Second Round |
| 7 | Rhode Island | 28 | 17.3 | Second Round |
| 8 | Missouri | 29 | 15.6 | Second Round |
| 10 | Butler | 30 | 13.7 | Second Round |
| 8 | Creighton | 31 | 13.5 | Second Round |
| 7 | Arkansas | 32 | 13.1 | First Round |
| 11 | UCLA | 33 | 12.0 | First Round |
| 8 | Virginia Tech | 34 | 11.9 | Second Round |
| 9 | Alabama | 35 | 11.9 | First Round |
| 9 | Kansas St. | 36 | 10.6 | First Round |
| 9 | Florida St. | 37 | 9.9 | First Round |
| 10 | Texas | 38 | 8.9 | First Round |
| 10 | Providence | 39 | 8.6 | First Round |
| 10 | Oklahoma | 40 | 8.1 | First Round |
| 11 | St. Bonaventure | 41 | 6.8 | First Four |
| 11 | Loyola Chicago | 42 | 6.6 | First Round |
| 12 | New Mexico St. | 43 | 5.8 | First Round |
| 11 | San Diego St. | 44 | 5.7 | First Round |
| 11 | Syracuse | 45 | 5.0 | First Round |
| 11 | Arizona St. | 46 | 5.0 | First Four |
| 12 | South Dakota St. | 47 | 4.4 | First Round |
| 9 | North Carolina St. | 48 | 3.9 | First Round |
| 12 | Murray St. | 49 | 3.7 | First Round |
| 14 | Montana | 50 | 2.8 | First Round |
| 13 | Buffalo | 51 | 2.3 | First Round |
| 14 | Stephen F. Austin | 52 | 2.2 | First Round |
| 13 | College of Charleston | 53 | 1.9 | First Round |
| 12 | Davidson | 54 | 1.9 | First Round |
| 13 | UNC Greensboro | 55 | 1.4 | First Round |
| 14 | Wright St. | 56 | 1.1 | First Round |
| 13 | Marshall | 57 | 1.0 | First Round |
| 15 | Georgia St. | 58 | 0.5 | First Round |
| 16 | Radford | 59 | 0.5 | First Round |
| 14 | Bucknell | 60 | 0.4 | First Round |
| 16 | Penn | 61 | 0.4 | First Round |
| 15 | Cal St. Fullerton | 62 | 0.4 | First Round |
| 16 | Texas Southern | 63 | 0.3 | First Round |
| 16 | UMBC | 64 | 0.2 | First Round |
| 16 | North Carolina Central | 65 | 0.1 | First Four |
| 15 | Iona | 66 | 0.1 | First Round |
| 15 | Lipscomb | 67 | 0.0 | First Round |
| 16 | LIU Brooklyn | 68 | 0.0 | First Four |
The model is very conservative in its early picks, probably due to the heavier weights assigned to scoring the entire tournament (there are variations on the model that score the First Round, Final Four, etc, alone, and we will be exploring these later). The model picks only one first round upset, 10-seed Butler over 7-seed Arkansas; even the 8s are all stronger than the 9s! Only one interloper makes the Sweet Sixteen—6-seed Florida. It's not until the Elite Eight, where top seeds Virgina and Xavier fall, that things get interesting.
So far the best "genes" are variations on our Strength power rating, several variants on Ken Pomeroy's data (dealing with tempo-corrected offense and defense especially), and the AP poll, both pre-season and final rankings. Also, we've found that for top teams, consistency is very important. We will continue to add new heuristics to the mix and see what works and what doesn't—or rather, which genes live and which ones die off.
Comments