Simple RPM Predictions for the NBA Conference Finals
By Nick Restifo
Feb 26, 2015; Cleveland, OH, USA; Cleveland Cavaliers forward LeBron James (23) drives between Golden State Warriors guard Klay Thompson (left) and guard Stephen Curry (30) in the third quarter at Quicken Loans Arena. Mandatory Credit: David Richard-USA TODAY Sports
(Ed: The NBA Conference Finals are upon us, and call for us bringing out the big guns. In this case Nick Restifo’s (@itsastat) Real Plus-Minus-based simulations of each series.)
The chance of the home team winning is evaluated by four different models, two different neural networks, a CRT decision tree, and a logistic regression, which are all trained on every playoff game since 2013-2014 (as far back as publicly available RPM goes), and are based on the minute-weighted RPM of each team. The minutes played average of each player during the second round of the 2015 NBA playoffs are used as their minute weighting for each predicted game, and the RPM used for each player is each player’s regular season RPM.
These predictions only take into account the weighted RPM of each team, and are trained on a very small amount of data, so they should be taken only as a ballpark of how the conference finals will go. They consider no other factor other than homecourt advantage, and there are many other factors that influence whether a game is won or lost.
Though the models used in the second round predictions were slightly different, (I believe that this model ensemble will yield a slight improvement in prediction accuracy), predicting via this procedure was somewhat successful (depending considerably on how you define success) in the second round. This methodology correctly predicted three of the four second round series to within a game, and although it incorrectly predicted the Clippers would beat the Rockets in 6 [1. ed: Clippers in 6 is awarded partial credit.], it predicted the series would go to seven games a sizable 29.26% of the time, and correctly ascertained that the Rockets would be 54.20% favorites should the series go to a game 7.
The prediction percentages listed here are based on 10,000 simulations of the conference finals, where the home team winning each game is decided by whether or not their chance of winning is over a randomly generated number between 0 and 1.
| Team | 4 Game Series Win% | 5 Game Series Win% | 6 Game Series Win% | 7 Game Series Win% | Win% |
| GS | 21.44% | 33.27% | 18.64% | 15.46% | 88.81% |
| CLE | 8.37% | 15.94% | 19.10% | 17.08% | 60.49% |
| ATL | 4.52% | 8.96% | 12.56% | 13.47% | 39.51% |
| HOU | 0.58% | 1.77% | 4.94% | 3.90% | 11.19% |
Cavs in 6 and Warriors in 5. Bring on these finals.