Fig. 3: Structural comparison of the networks.
From: Forecasting influenza activity using machine-learned mobility map
a Correlation between networks. Pearson correlation coefficient between the different mobility networks (FIPS-sorted by source and destination). AMM and COMMUTE have a high positive correlation value (0.90), followed by RADIATION model (0.71 with AMM), while GRAVITY has relatively small positive correlation with the other networks. b–e Adjacency flow matrices of AMM, COMMUTE, GRAVITY, and RADIATION. Nodes are arranged by spatial proximity. Heatmap color is indicative of normalized flow volume (darker color denotes larger flow). Note that while AMM and COMMUTE are sparser and seem to be clustered, GRAVITY network, by virtue of its definition, is more homogeneous. More detailed versions of AMM and COMMUTE flows are shown in Supplementary Figs. 3 and 4. f Betweenness measures at county level for all four networks. Among the four networks, AMM, and RADIATION models have similar sets of top counties by betweenness, while COMMUTE chooses nearby but distinct counties. These counties are reasonable choices, since they connect rural counties, e.g., Allegany or St. Lawrence or Cape May, to highly urbanized counties like Manhattan or Brooklyn. However, in GRAVITY we failed to identify any such distinct pattern.
