Inspiration
The inspiration for this NBA data model stems from the desire to gain a competitive edge in the fast-paced world of professional basketball. The model aims to predict whether the home team will win a game, not only before it starts but also during the game itself12. This real-time prediction capability provides valuable insights for coaches, analysts, and fans alike, revolutionizing basketball strategy and engagement.
What it does
It predicts who will win the NBA game Challenges we ran into
- Multiple important features were of different types, requiring careful handling. We faced RAM limitations, forcing us to reduce the amount of data we could use for training. Using the rankings posed a challenge because early in the season, if a team has a record of 1-0, their win-ratio will be 1.0, which can skew the results. We fixed this by removing data for the first three games of each team.
Accomplishments that we're proud of
1) To improve model accuracy by increasing relevant features
What we learned
1) How to build a regression model, feauture building 2) How to align snowflake, DBT tranformations
What's next for Hoops and Algorithms: Predicting NBA winners
Our current model imitations in our current model: Computational constraints that limit the amount of data we can process The complexity of capturing all relevant factors in a fast-paced game For future work, we plan to: Be able to use our models and scrape live games in order to predict the winner in a game that is currently happening Incorporate more advanced features, such as player matchup data and team-specific strategies
Log in or sign up for Devpost to join the conversation.