You are here

Can Machine Learning and Crowdsourcing Fight the Flu?

College researchers are exploring the most accurate methods to predict flu season.

Thanks to cutting-edge research from Carnegie Mellon University, artificial intelligence may soon be able to alert public health facilities when flu season will be at its worst.

CMU’s Delphi research team, a blend of students and faculty from the university’s department of machine learning, statistics, computer science and computational biology, is part of a Centers for Disease Control initiative investigating how to accurately predict weekly flu season forecasts.

“We’re using multiple approaches, most driven by machine learning,” says Roni Rosenfeld, a computer science professor who heads up the team. “The methods use large amounts of data and statistical models to make predictions.”

The Delphi team, which uses machine learning “to make predictions based on past patterns and on input from the CDC’s domestic influenza surveillance system,” boasted a forecast last year that was within 25 percent of the CDC’s best estimate about 75 percent of the time, an article on CMU’s website reports.

Human Predictions vs. Artificial Intelligence Analysis

Surprisingly, a partially crowdsourced prediction method was one of CMU's most successful. In a design presented by researcher David Farrow, this “human method” involved a group of volunteers logging onto a website called Epicast, where they had access to informational resources like past CDC data, to make a prediction about the number of flu cases.

“If many people were to make such predictions, would the aggregate forecast be accurate?” writes Farrow in his research thesis.

The answer to this question has been yes. The CMU article reports that this human system was the top-ranking forecasting system for the 2014-2015 flu season — the AI system took the top spot last year.

“People are very adept at adjusting to different circumstances and still make a reasonable estimate,” says Rosenfeld. “They may not be as accurate as the computer methods, but they are less likely to make big mistakes.”

Whether forecasted by machine, humans or a combination of the two, the research work being done at CMU has the potential to help streamline the treatment of the flu at hospitals and college medical centers.

“I think the applications are enormous,” says Rosenfeld. “For healthcare and government public health agencies, knowing when the largest demand for treatment will come is huge.”

Mar 16 2017