INDIANAPOLIS—Everybody knows when the Golden State Warriors are in town.

There’s a buzz around the arena earlier than usual, as fans make sure to arrive at least 90 minutes before tipoff to catch a glimpse of Stephen Curry’s famous pregame ritual, complete with two-ball dribbling drills and circus-like shots from the tunnel. All of a sudden, a random Wednesday night in name-that-NBA-city is not so mundane. And game No. 24 on the drawn-out NBA schedule is not so meaningless. It doesn’t take a rocket scientist to realize when the winners of three of the last four NBA titles comes to town, attendance will be up.

But, what about all those other random games packed into an 82 game NBA season? When Curry isn’t there to get the crowd excited? When the weather is bad? Or when the average person cannot name a single player on the opposing team? Turns out there is a way to predict attendance on those nights, too.

Enter Butler University Associate Professor of Operations Management Barry King. And enter his algorithm-based, machine learning approach to NBA attendance predictions.

“We were able to predict attendance by looking at home team popularity, Twitter followers, day of the week, home team winning percentage, home city’s total personal income, and other variables,” says King, who specializes in predictive analytics. “By taking those predictor variables, along with historical data, we were able to come up with an accurate forecast that can have many applications beyond just the NBA.”

King’s findings were certainly accurate. Published in the Journal of Computer Science & Information Technology, King was able to predict attendance at every regular season game from the 2015-2017 NBA seasons, on average, within five percent.

To get an accurate prediction, King explains, he trained a type of algorithm (Random Forest) to predict an outcome using historical data. This, he says, is machine learning. Machine learning leverages historical data to inform future forecasts.

So, King trained the machine. He plugged in attendance data from the 2009-2013 NBA seasons into the algorithm, along with predictor variables like home team popularity, popularity of the opponent, day of the week the game occurred, home team winning percentage, home city’s total personal income, and capacity of home venue.

“We are among the first to use machine learning to predict attendance,” he says. “That is unique because it takes historical data into consideration. We believe that training the machine on historical data enabled us to get a much more accurate prediction of future attendance. Taking history into account, and teaching the machine that history, enables the machine to come up with future forecasts.”

King has applied this method of predictive analysis to the NHL and MLS. And the accuracy remained. While he now has the ability to accurately predict the attendance for these professional sport leagues, he believes the application goes beyond the wide world of sports.

“This has carry over to the business world and how companies can run their enterprises better,” he says. “As a manager of a basketball team, I would certainly like to know how many people are likely to show up for a random February game so that I can plan to have more staff on hand, if needed, or start to think about amping up the promotions, if attendance looks low. This could also help teams determine ticket price levels.”

Machine learning, King says, is an important area when it comes to forecasting. In the future, he says, he would like to build his prediction tool into an app that industry people can use to easily access this information, and then make decisions based off the results, on their own.

King says the information can also be applied to coming up with scheduling at a hospital, crews on airlines, and those are just some examples.

“Real world solutions often start with having a good idea of what the future might look like,” he says. “We now have a way to make accurate future predictions, based on historical data. I see this being useful in many industries.”