Like many University of Virginia students, David Lorenz filled out a bracket and rooted for the Virginia men’s basketball team to win it all in the recent NCAA Tournament. His pick of Virginia to cut down the nets in Minneapolis was only partially due to school spirit, though.
Rather, Lorenz applied the machine-learning tools he’s using in UVA’s M.S. in Business Analytics Program to build a model to predict winners of each game in the tournament.
Lorenz’s data-driven prognostications were accurate enough to take fourth out of 868 entries in the Google Cloud & NCAA Machine Learning Competition, an annual event in which participants use reams of historical NCAA data to build predictive models. Lorenz’s winning entry, which predicted 76% of the games correctly, came with a $2,000 prize.
“I had an idea that I had a good model because I knew it performed well in the test set,” Lorenz said, referring to the previous years’ completed tournaments. “I had a feeling the model I built was going to do well in the competition, but not necessarily as well as it did.”
Lorenz, a Senior Consultant at the economic consulting firm Bates White, said he combined historical regular-season data, tournament data, and metrics from leading basketball statisticians, and then evaluated and tweaked the model based on a variety of parameters.
The most effective model was a neural network that included 25 variables, ultimately leading to a successful prediction of the Cavaliers winning it all.
“To get into the top 10%, it takes a good model, but top 5 overall has a luck factor,” Lorenz said, noting, for instance, that he wouldn’t have placed nearly so high had Texas Tech pulled out the victory in the thrilling overtime championship game.
One factor in Lorenz’s favor: He had already completed a similar task in one of his MSBA courses. In a class taught by Darden School of Business Professor Casey Lichtendahl, Lorenz and classmates built predictive models using prior tournament data. For the Google-sponsored challenge, Lorenz collected additional historical data and sources but went in with a sense of how to build an effective model.
Basketball predicting acumen has been an unexpected bonus in the MSBA Program, a 12-month degree program delivered in partnership by the McIntire School of Commerce and Darden School of Business, which Lorenz enrolled in with the goal of learning more about how the application of data could help solve business problems.
“I can’t speak highly enough about the program,” said Lorenz. “I knew I would be learning a lot, but didn’t realize the quality of the professors at UVA and the quality of my peers.”
Lorenz said he expected to be working hard and developing his analytical skills, and both expectations have been met. The biggest surprise, he said, is how much he enjoys attending class.
By Dave Hendrick
Learn more about the UVA MSBA Program.