Data Competition Upgrades Zestimate Accuracy
The ability to reckon the value of what for many is their most valuable asset increased with the award this week of the Zillow Prize, a contest to tweak the accuracy of its Zestimate algorithm.
The winner of the $1 million prize, an international team of data scientists and engineers from Canada, Morocco and the U.S., topped the Zillow benchmark model by roughly 13 percent. The improvements mean Zestimate’s current national error rate for estimating home value will drop to less than 4 percent, a half-a-percent improvement.
(Three years ago, the company pegged its median national error rate at 8.3 percent.)
The winning team included Nima Shahbazi of Canada, Chahhou Mohamed of Morocco and Jordan Meyer of the U.S. Zillow said Wednesday (Jan. 30) its data scientists will now incorporate parts of the winning team’s work into the Zestimate model that gauges the valuations of about 110 million U.S. homes.
Zillow’s estimated market value is based on an array of public and user-submitted data, including location, lot size, square footage as well as number of bedrooms and bathrooms. Historical data like real estate transfers and tax information are also factored in, as are sales of comparable houses in a neighborhood.
The winning team’s algorithm incorporated emerging machine learning techniques, including deep neural networks used to directly estimate home values and “remove outlier data points that fed into their algorithm,” Zillow said. They also used public data such as rental rates, commute times and home prices along with “contextual information” such as road noise.
The improved algorithm is expected to bring future home value estimates about $1,300 closer to actual sale prices.
“We’ve been on a 13-year journey making the Zestimate more accurate, and hosting Zillow Prize allowed us to invite thousands of brilliant data scientists from around the world to join us on this journey,” said Stan Humphries, Zillow’s chief analytics officer and creator of the Zestimate.
The Zestimate competition was launched two years ago, quickly becoming the most popular machine learning competition on the Kaggle data science platform. The second-place team won $100,000 while third place was worth $50,000.
In announcing the competition, Seattle-based Zillow said itmarked the first time a part of its proprietary data behind the Zestimate home valuation would be available to outside researchers.
Winning team member Nima Shahbazi, CEO of a machine learning startup aimed at the financial sector, said he may use part of the prize money to invest in real estate.