How Google Can Get the Flu Right

An anonymous reader writes Google Flu Trends was developed in 2009 to improve forecasts of flu levels in the U.S. by utilising Google search data. This early example showcased the potential which lies in the exploitation of human digital traces which all of us leave behind by using online services. The rise of Google Flu Trends was only stopped when the service dramatically overestimated the number of flu incidences recently. The fall raised questions about the value of online data for predictions in general. However, a study published yesterday demonstrates that it is not only about data but also about the adaptiveness of algorithms used for predictions. Scientists combined historic flu levels as reported by the CDC with Google Flu Trends data using an algorithmic framework which is able to adapt to changes in human search behaviour. Their results show that Google Flu Trends data sets significantly add information to the forecasts of current flu levels.

Comments

Indirect measurement

[Dan East]

This is a very indirect way to measure and thus predict flu outbreaks. There are many things that could prompt people to search for and otherwise discuss the flu, beyond a person or their family actually having the flu. All it would take is a few major news sources to make mention of the flu, and there's a good chance there would be a surge in people searching for terms related to the flu. Further, other illnesses could be going around that have some flu-like symptoms which would also increase search results for the flu. Can't we just get this information from healthcare providers or insurance companies directly? (although the latter is probably charging an arm and leg for that information)