Using Machine Learning to Predict NBA Outcomes and Applying Portfolio Theory and the Kelly Criterion to Optimize Betting Profitability

Abstract

In 2022 the sports betting market grew by 75%. The federal ban on sports gambling was lifted in 2018 and it has become legal in 19 states in 2020 and 33 states in 2023 [1]. The growing market may also present opportunities at beating sports gambling websites to generate a profit. I created a ridge regression model to try and predict the winners of NBA games. I eventually refined the model until I got an accuracy score of 63%. While at first I wanted to also look at bet size optimization as most research on this topic has been in either creating more accurate predictive models to beat existing the sports lines or optimizing a portfolio of bets. I would like to combine the two, creating a betting strategy using Markowitz portfolio theory and the Kelly criterion beat the sports betting market based off a machine learning model. However as I worked, I realized that bet size optimization would have to wait as a worked on improving my models as they would be pointless if my predictions were way off.