Closing out a nearly two-year-long data science competition, Zillow Inc. awarded a $1 million prize Wednesday to a loosely federated group of three people who figured out how to beat its home price-estimating benchmark model by 13 percent.

Zillow said the techniques will reduce the company’s Zestimate’s current nationwide error rate of 4.5 percent to below 4 percent.

Zestimate is an estimated market value of homes that Zillow computes using a proprietary formula that computed daily based upon millions of public and user-submitted data points. Currently, the Zestimate usually falls within $10,000 of a home’s sale price; Zillow expects the improvements could bring it $1,300 closer.

The winning team among 3,800 from 91 countries, Team ChaNJestimate, was comprised of Chahhou Mohamed of Morocco, Jordan Meyer of the U.S. and Nima Shahbazi of Canada. Nima and Chahhou had initially competed as a team during the contest’s early stages while Jordan went solo.

The three joined forces after realizing the potential of combining their models, Zillow said. They still haven’t met in person but have spent hundreds of hours collaborating virtually.

“It’s amazing to know that millions of people will benefit from our ideas,” said Shahbazi in a prepared statement. “For every idea that worked, there were a hundred that didn’t work. But we kept going.”

With data scientists in short supply and salaries skyrocketing, competitions have become a popular way for organizations to find creative solutions to their big-data problems quickly and cost-effectively. No one knows the size of the data science competition market, but Google LLC-owned Kaggle and Topcoder Inc., which operate the two largest competition platforms, collectively boast nearly 1.5 million members.

The winning team’s algorithm incorporated both machine learning and deep learning techniques. They used neural networks to estimate home values more accurately by removing outlier data points. They also leveraged publicly available external data including rental rates, commute times, home prices and even factors such as road noise that influence sales price.

Zillow has already begun to incorporate parts of the winning team’s algorithm into its model, which covers 110 million homes across the U.S.  The contest was run on Kaggle’s competition platform. Zillow also awarded $100,000 to the second-place team and $50,000 to the third-place team.

Image: Zillow