NBA Player Potential Predictions

NBA teams spend hundreds millions of dollars on extensions for young players. Some players might ascend immediately while others breakout later. We aim to explain what indicates true potential.

Comment

Inspiration

Throughout the NBA draft cycle, trade negotiations, and free agency, analysts are always discussing a players floor and ceiling. We wanted to know if there was a way to determine the range of outcomes of how good a player could truly be, and what factors decide it.

What it does

Our application predicted historical peak season stats(points,rebounds,assists) and users can view these predictions compared to the actual results. We also predicted current young players with 3 years of experience who have yet to reach their peak.

How we built it

Using a data set from Kaggle we were able to get historical data from (1998-2025) that included all relevant statistical categories. Then we cleaned the dataset to remove any irrelevant or incomplete data points. Next, we explored that data to find patterns and determine correlations. We moved from naive models like linear regression to more complex models such as Random Forest, and Ridge.

Challenges we ran into

Feature selection was challenging since there were many variables that either didn't contribute to model or created unnecessary 'noise' skewing the overall results. Furthermore, there are many player archetypes across positions that make it difficult to generalize.

Accomplishments that we're proud of

We deployed a live web application and our error was < 3.0 for points and < 1.0 for assists/rebounds.

What we learned

When working with an AI platform like Zerve AI while you can move from ideas, to exploration, to results very quickly you must ensure that you have proper guardrails.

What's next for NBA Player Potential Predictions

We plan to include more biometric data and more in-game metrics for more accurate predictions. Hopefully, in the future, we would be able to apply similar methodology to other sports.

Built With

Share this project:

Updates