Feature Selection for Real-time Estimates of Influenza-like Illness

Influenza epidemics causes significant health issues and economic burden to the society. It is estimated that the annual epidemics result in 3 to 5 million cases of severe illness, and about 250,000 to 500,000 deaths worldwide (WHO, 2016). Economically, the influenza epidemics contribute a $87.1 billion burden on the United States (US) alone each year (Molinari et al., 2007). One way to reduce the losses is to provide accurate, reliable forecasts of the influenza epidemics. Short-term forecasts within a season can help policy makers tailor the vaccine campaign, and guide individuals and organizations to adjust their activity plans to curb the influenza transmission. Long-term predictions can provide valuable information on selecting vaccines for future seasons (Brooks, Farrow, Hyun, Tibshirani, & Rosenfeld, 2015). With well-documented surveillance data for influenza-like illness, and various other digital surveillances such as search engine and social network data available nowadays, these goals seem feasible.