Figure 5
From: Finding disease outbreak locations from human mobility data
Robustness of the method to noise in the input data and secondary infections. A We apply Gaussian noise with standard deviation \(\epsilon _{\mathrm{noise}}\) to the location data of \(N=4\) individuals. The accuracy decreases only after unrealistically high levels of noise are applied, compared to the mean smartphone GPS accuracy of \(4.9~\mathrm{m}\) [63]. B We test the robustness to individuals in the sample that were not present at the outbreak origin, for example due to secondary infections. Starting with a sample of \(N=10\) individuals, we replace a fraction of \(r_{\mathrm{noise}}\) trajectories with random individuals from the dataset. The inference method is robust to high amounts of noise or secondary infections for the synthetic data, and moderate amounts for the empirical datasets