NBA Postseason Probability Model

AE4three
Registered User
Joined: 02/04/2012
Posts: 21
Points: 36
Offline
NBA Postseason Probability Model

I've created and posted an NBA Postseason Probability Model onto my website. Check it out if interested....


Striver347
Registered User
Joined: 12/31/2009
Posts: 64
Points: 300
Offline
Very neat work. I'm also in

Very neat work. I'm also in the Data Science & Machine Learning space. I followed your GitHub and I'll check out some of your repositories soon!

Arc12345
Registered User
Joined: 07/13/2009
Posts: 494
Points: 1699
Offline
Timberwolves

So if I'm understanding this correctly, you have the Timberwolves with the 9th highest probability to make the playoffs?

If that's the case I don't like this math. Incredibly cool though.

StephGoneSteph
StephGoneSteph's picture
Registered User
Joined: 05/17/2016
Posts: 78
Points: 341
Offline
cool but some fishy stuff

Some of the probabilities seem incredibly high. If i'm reading this correctly, it says the Timberwolves have a 96% chance of making the playoffs? That seems incorrect, given that they wouldn't be in the playoffs if they started today. You say the Pelicans have an 88.5% percent chance of making it, despite being 13th in the WC standings. What statistics are driving the results for these two teams in your model?

I guess the sum of probabilities should be about 8 for each conference. But the sum of probabilities in the West is 11.

StephGoneSteph
StephGoneSteph's picture
Registered User
Joined: 05/17/2016
Posts: 78
Points: 341
Offline
Also the Pelicans are in the

Also the Pelicans are in the West. You have them in the East on your bar graph.

StephGoneSteph
StephGoneSteph's picture
Registered User
Joined: 05/17/2016
Posts: 78
Points: 341
Offline
Also the Pelicans are in the

Double post sorry

sweaterflex
sweaterflex's picture
Registered User
Joined: 05/12/2009
Posts: 364
Points: 1476
Offline
Good note

This actually has a really easy fix in the model, the model should be split into Eastern Conference playoff probability and Western Conference probability and then joined post-calculation for the visual. The data is trained on historical data for both conferences, and a lot of teams in the West would have 'historically' made the playoffs in more balanced eras. You lose a little bit of predictive strength due to smaller sample, but the current predictions will be much more accurate.

OhCanada-
OhCanada-'s picture
Registered User
Joined: 05/08/2010
Posts: 10454
Points: 14249
Offline
English please.

English please.

AE4three
Registered User
Joined: 02/04/2012
Posts: 21
Points: 36
Offline
the model...

The data the model was built on are end-of-season basic team and opponent statistics from years past (described on the site). I also included mean age and current win % (because it was pretty easy to get from basketball-reference.com). The model is going to predict correctly about 90% of the time (see confusion matrix and classification report) which means it will probably miss on a few teams each year, but that is doing incredibly well given the limited amount of data included in the model. Please keep in mind the error bars which communicate that the probability is 95% likely to fall within that range.

StephGoneSteph
StephGoneSteph's picture
Registered User
Joined: 05/17/2016
Posts: 78
Points: 341
Offline
Sure

But the fact that the sum of probabilities is so far off suggest there's pretty big room for improvement. I'm not sure using data from the 70s is going to help you much, given that the number of total teams and number of playoff teams have both changes. Therefore, the relationship between current win %, for example, and probability of making the playoffs are fundamentally different.

Even looking at the bottom of your error bars- the bottom of the error bars for the Timberwolves, Jazz, Mavericks and Pelicans are all above 50. Therefore you easily reject that any of these teams have less than a 50 percent chance of making the playoffs. A cursory glance at the current standings tells you that can't be the case.

AE4three
Registered User
Joined: 02/04/2012
Posts: 21
Points: 36
Offline
misses on 4/30

Right, so if it misses on 4 teams (wolves, jazz, mavs, pelicans) in 30 thats about an 87% clip which is in-line with how it performed on the test data (~90%). Also, the season is still only half way over. Every model is a work in progress and this model is in its infancy.

StephGoneSteph
StephGoneSteph's picture
Registered User
Joined: 05/17/2016
Posts: 78
Points: 341
Offline
Sure

Agreed. Again, I thinks its cool. Just trying to offer some constructive feedback.

AE4three
Registered User
Joined: 02/04/2012
Posts: 21
Points: 36
Offline
no offense taken...

I appreciate the feedback and keep an eye out for updates to the model. I'll be changing it and hopefully improving it in the near future.

Zouldiers
Zouldiers's picture
Registered User
Joined: 10/27/2013
Posts: 707
Points: 1942
Offline
5% error or 1% error?

5% error or 1% error?

frogman
Registered User
Joined: 06/07/2011
Posts: 1049
Points: 2988
Offline
In the Western conference you

In the Western conference you have every single team except the Suns above a 50% chance of making the playoffs. That doesn't really check out.

RSS: Syndicate content