Inspiration
Wanted to learn about different AI models and incorporate into something we care about.
What it does
Predicts the outcome of NBA games based on the previous three seasons worth of games.
How we built it
Next_js for frontend, Django + nba_api for backend.
Challenges we ran into
Wanted to incorporate individual player statistics (e.g. how a player does versus a specific team compared to season averages), however that means we would need to get a seasons worth of all player statistics for each team. That comes out to 30 teams and 82 games, so 30 * 82 = 2460 API calls. We do not have access to that.
Accomplishments that we're proud of
Our model is at a 68.5% accuracy. With the resources that we were given, our goal was to reach 65% as simply guessing off of whatever team has the better win percentage results in a 59% accuracy. So we started at 59% and slowly worked our way towards 68.5%.
What we learned
Sometimes adding too much data actually decreases the accuracy. This happened when I added pace as a feature.
What's next for HoopPredict
Find a way (maybe a different API or another method) to get player statistics per game so we can get our model to become even more accurate.
Log in or sign up for Devpost to join the conversation.