AGRONet leverages its information from electronic health records, flu-related Google searches and historical flu activity in a given location. Improved accuracy has been achieved by adding a second model, which draws on spatial-temporal patterns of flu spread in neighbouring areas . Furthermore, the machine learning system was “trained” by feeding it flu predictions from both models as well as actual flu data, helping to reduce errors in the predictions.
Figure 1: State-level performance of ARGONet benchmarked with ARGO and AR52, as measured by RMSE (top), Pearson correlation (middle), and MAPE (bottom), over the period from September 28, 2014 to May 14, 2017. States are ordered by ARGONet performance relative to the benchmarks to facilitate comparison 
The novel methodological framework for this task dynamically combines two distinct flu tracking techniques, using ensemble machine learning approaches, to achieve improved flu activity estimates at the state level in the US. The two predictive techniques behind the proposed ensemble methodology, named ARGONet, utilize (1) a dynamic and self-correcting statistical approach to combine flu-related Google search frequencies, information from electronic health records, and historical trends within a given state, as well as (2) a data-driven network-based approach that leverages spatial and temporal synchronicities observed in historical flu activity across states to improve state-level flu activity estimates.
“It exploits the fact that the presence of flu in nearby locations may increase the risk of experiencing a disease outbreak at a given location”
– Harvard Medical School.
Read more here.
Written by Nefti-Eboni Bempong