Predicting the NBA Playoffs

Explanation

Here is how I attempted to predict the upcoming NBA Playoffs bracket using Machine Learning. I began by creating the following five models:

1. WinLoss% Logistic Regression

   • Both teams WL% onto series winner

   • Yielded 0.6732 accuracy rate

2. ELO Model

   • Series simulated using single game win probability formula:

     • P(A) = 1/(1+10^m); m = (Rating(A)-Rating(B)/400

3. Monte Carlo Simulations

4. Neural Network

   • Several offensive and defensive stats as inputs

   • Yielded 0.7722 accuracy rate

5. Random Forest

   • Same offensive and defensive stats as inputs

   • Yielded 0.6930 accuracy rate

I performed stacking on the five models to calculate overall win probabilities. The stacked model had a 0.86633 accuracy rate on the test data. You will notice that many of the models clearly prefer the higher seeded, home team. Stacking the models helped combat this and predict more upsets.

Unfortunately, injuries are not accounted for in these predictions. Injuries to players like Giannis, Kawhi, and Jimmy Butler are certainly significant and would alter the odds of victory.

PREDICTIONS

Here is the models predicted playoff bracket:

*Grayed out teams are incorrect predictions

Down below is the models breakdown of each series that will actually occur.

FIRST ROUND

EASTERN CONFERECE

#2 Knicks vs #7 76ers

#1 Celtics vs #8 Heat

Pick: Celtics

#3 Bucks vs #6 Pacers

Pick: Pacers

Pick: Knicks

#4 Cavaliers vs #5 Magic

Pick: Magic

WESTERN CONFERENCE

#1 Thunder vs #8 Pelicans

Pick: Thunder

#3 Timberwolves vs #6 Suns

Pick: Timberwolves

#2 Nuggets vs #7 Lakers

Pick: Nuggets

#4 Clippers vs #5 mavericks

Pick: Clippers

SECOND ROUND

6/8 First Round Series Predicted Correctly

Pre-Second Round Probabilities

EASTERN CONFERENCE

#1 Celtics vs #4 Cavaliers

#2 Knicks vs #6 Pacers

Pick: Celtics

Pick: Knicks

#2 Nuggets vs #3 Timberwolves

#1 Thunder vs #5 Mavericks

Pick: Thunder

Pick: Nuggets