Winners and Whiners: Applying Deep Learning to the NBA

Winners and Whiners: A Deep Learning Model to Predict NBA Outcomes

Introduction Sports betting has become a very popular trend within the world of the young sports and gambling community. There are now thousands and thousands of different ways one can gamble or bet on a professional sports game, from Korean Basketball to F1 Racing. With this new phenomenon, people have turned to data science and deep learning to better understand the results of individual games and how they can take advantage of odds that are, in theory, “mis-priced”. Deep learning algorithms offer much value to this topic and ultimately allow for the analysis of the question “Can sports games be predicted to a certain degree?”. To better understand this, we decided to use NBA data to build a model that would predict final scores between two teams, as well as the ultimate winner within a specific confidence interval. We also implemented play by play data to build a standard sequential LSTM that predicts the winner of an NBA game based on sequential play by play data of a given game. In theory, this would open up avenues for live-sports betting algorithmic work as a play-by-play model could ingest a certain amount of plays during a live game and try to predict outcomes. Ultimately, the goal of our project was to analyze NBA data and statistics and use several deep learning techniques and algorithms to predict outcomes of games with an accuracy above 50%.

Models: Pre-Game Model: Feed Forward Neural Network Play-By-Play Model: LSTM, LSTM + CNN

Things we Learned: Data processing is hard! And applying Deep Learning to our passion of the NBA was awesome!

Code: https://github.com/polleyethan/winners_and_whiners

Final Writeup: https://docs.google.com/document/d/1WDhPd1533yLVNhhfSyN4Dyo-WXumI3-pvQvP4ivzPg0/edit?usp=sharing

Reflection 2: https://docs.google.com/document/d/1Wzpb9dojqtj-k8b_WdnkfAQdOOIzuQAWChU5QgxzF3A/edit?usp=sharing

Reflection 1/Proposal: https://docs.google.com/document/d/1Dl2wCekjjXbIksmDWniczKMouXoVtCJZrxWJSK96qQ0/edit?usp=sharing