|Google senior software engineer Christian Stefansen|
Google ($GOOG) has made the biggest change to its Flu Trends influenza tracking software since it first released the system in 2008. The update incorporates data from the Centers for Disease Control and Prevention (CDC) in an attempt to improve the accuracy of the model.
Flu Trends has been through a topsy-turvy period, starting in 2009 when its ability to predict the H1N1 outbreak before traditional influenza trackers saw it heralded as a new, smarter way of monitoring disease. The fall came in the 2012-13 flu season, when Flu Trends overestimated the incidence of influenza. Mountain View, CA-based Google tweaked the model for last year's flu season--and got more accurate results--but has now decided a bigger overhaul is needed.
Google provided news of the changes on its blog, but is withholding technical details of the revisions for an upcoming paper. On the blog, Google senior software engineer Christian Stefansen outlined how Flu Trends in the U.S. will now factor official CDC data into its calculations as the influenza season progresses. The addition of CDC data ends Flu Trends' complete reliance on search terms and also allows Google to update the model as the influenza season progresses.
In the past, Google tweaked the model between flu seasons but didn't do real-time updates. The lack of feedback-driven improvements during the flu season meant the model could veer wildly off course, as happened in 2012-13. Having the model respond to CDC data should mean it recalibrates its forecasts when there is a major discrepancy. Several academic publications--including the Science paper that picked apart the 2012-13 failings--have found that adding CDC data would improve accuracy.