Using machine learning to predict the 2019 MVP and All-NBA teams: end of season predictions

There are summary infographics at the bottom if you want to skip to the results.

Over the past few months, we’ve been using machine learning to predict the MVP (first post with methodology, most recent update) and All-NBA teams. The models made new predictions for these most recently in February.

Since then, we’ve seen lots of risers and fallers. Though the MVP race remains a two-man discussion between James Harden and Giannis Antetokounmpo, the remaining MVP contenders changed, as did the All-NBA teams.

For example, at the time of the most recent predictions post, Paul George was in the midst of an incredible scoring run which lead the Thunder to the third seed in the West and the sixth-best record in the league (37-20 before the All-Star break). However, in the months following the break, George saw drops in his shooting due to a shoulder injury.

Before the All-Star break, George averaged 28.7 PPG on an excellent 40.6% from three. However, post-All-Star break, his PPG dropped by more than 2 to 26.4 and his three-point percentage dropped seven percentage points to 33.6%.

This combined with the Thunder’s drop to the sixth seed not only prevents him from being a clear third place in the MVP race but also jeopardizes his once-clear spot on the first team after Giannis. Kevin Durant’s absurd efficiency lately puts him in a close contest with George for that spot.

Though George underwent a particularly large shake-up, several players saw changes affecting their All-NBA status. Harden, Curry, and Giannis are still locks for the first team, but lots of spots remain up for grabs.

Embiid’s injuries throughout the season may prove to be the difference between him and Jokic for the first team center slot. We could also see some changes in the third team; at the time of the initial post, Ben Simmons got the final guard slot with only a slight advantage over Kemba Walker and Bradley Beal.

Let’s use the same models we used to predict the MVP and All-NBA teams earlier in the season with complete data now that the season is over and see who deserves what award.

Methods

Because previous posts discussed the methods for both models extensively, we’ll discuss them minimally.

Our model predicting the MVP takes the factors in the table below to predict vote share (note that vote share = % of maximum votes, so it doesn’t add up to 1).

The player with the highest predicted vote share is our predicted MVP. To read more about the models and gain a more in-depth view of the model, view the initial post here.

Our model predicting All-NBA teams is a bit different. While the MVP model is a regression problem where we’re predicting a numerical value between 0 and 1, All-NBA prediction is a classification problem. We’re essentially predicting if a player is All-NBA caliber. However, by looking at the prediction probabilities – the model’s certainty that a player is All-NBA caliber – we can assemble teams with the correct number of players in each position. The player with the highest prediction probability is placed into the highest available slot for his position.

The All-NBA model takes into account the same factors as the MVP model (in the table below). However, it also considers whether the player was an All-Star or not – players who made an All-Star team have a value of 1, while those who don’t have a value of 0.

To learn more about the All-NBA models, view the initial post here.

In both cases, four models were made: a support vector machine, random forest, k-nearest neighbors, and deep neural network.

We’re using the exact same models, just with data from the end of the season. This will give us final predictions for the MVP and All-NBA teams.

Results: MVP

The four graphs below show each model’s MVP predictions.

Two models predict Harden will win MVP, and two predict Giannis. This is quite an impressive late-season push by Harden, as Giannis had a sizable lead in both of the previous predictions. The two graphs below show the change in predicted vote share from today to the All-Star and mid-season predictions.

The graph below shows the average prediction from the four models.

Giannis and Harden are incredibly close; Giannis has an average predicted vote share of 0.676, while Harden has a vote share of 0.656. This comes out to a lead of just about 3% for Giannis.

Your 2019 MVP, by a very narrow margin, is Giannis Antetokounmpo.

Results: All-NBA

The four tables below show each model’s predicted All-NBA teams with the prediction probabilities in parentheses. Players with an asterisk have an identical prediction probability to another player with an asterisk on another team.

The table below shows the average of the four models.

The average predictions had two slots that were very close between the first and second teams. Steph Curry had an average prediction probability almost identical to Lillard’s (though slightly smaller). However, this spot is almost guaranteed to go to Curry.

Similarly, Jokic had a very small deficit of only 0.003 to Embiid for the first team. Given Embiid’s 64 games played compared to Jokic’s 80, this spot is likely to go to Jokic.

Compared to last time’s predictions, there were a few differences. Along with these shifts (Curry-Lillard and Jokic-Embiid) with tiny differences, Kevin Durant overtook Paul George for the second forward slot on the first team. However, narrative may deter voters from placing him on the first team.

The final change occurred in the last guard spot on the third spot. Previously, Ben Simmons had a small lead over Kemba Walker and Bradley Beal. However, the average predictions give the nod to Walker.

Conclusion

For our final NBA awards predictions of 2019, lots of spots became closer and closer to a toss-up. In the MVP race, the predicted vote shares for Giannis and Harden show an impossibly close race. Of the four models, two favor Giannis and two favor Harden. In the average of the models, Giannis has a lead of about 3%.

On the All-NBA front, the biggest change seems to be that Kevin Durant has eclipsed Paul George for the first team forward slot alongside Giannis. However, the race is close, and it could go to either player.

Infographics and summary tables

MVP results:

All-NBA results: