Building a Predictive Model for the NBA

The National Basketball Association is an economy that runs on data analysis. Every organization within the league has shifted their primary focus to data analytics for guidance with regards to most critical decisions. Each team evaluates players using their own models such that every player has a unique EV (expected value) and ROI (return on investment) to each team. Furthermore, the entire sports betting industry revenue has increased 44.5% in the last year thanks to a record high 119.84 billion dollars wagered by Americans in 2023 - a 27.5% increase in betting volume from the previous year (2022)¹. So it is not only NBA team organizations that are battling to build the best predictive models, but also sportsbooks all around the world. Vegas casinos have created algorithms that have made them billions of dollars. The ironic part is that over the past five NBA seasons, only 67.01% of the teams Vegas has picked to be the favorites have actually won. Although guessing more than 2/3 of NBA of matchup predictions correctly is impressive, it can shine light onto how difficult this task truly is². These larger companies can account for millions of data points to create these predictions, and yet still fall short by about 33% accuracy. Thus, I became curious to see if I could use free public data available to build my own comparable predictive model using the skills I have learned from both computer science and economics courses I have taken for my major.