Table 7. Classification of HIV trees into simulated trees from outbreaks with different β.
| HIV test clades | KNN | Random forest | ||
|---|---|---|---|---|
| β = 0.05 | β = 0.2 | β = 0.05 | β = 0.2 | |
| trained on: trees from static skewed-clustered networks | ||||
| all 90 | 0.63 | 0.37 | 0.66 | 0.34 |
| with >50% NL-tips | 0.23 | 0.77 | 0.08 | 0.92 |
| with >70% NL-tips | 0.12 | 0.88 | 0 | 1 |
| with <30% NL-tips | 0.71 | 0.29 | 0.75 | 0.25 |
| trained on: trees from dynamic skewed-clustered networks with δ = 0.1 | ||||
| all 90 clades | 0 | 1 | 0 | 1 |
| with >50% NL-tips | 0 | 1 | 0 | 1 |
| with >70% NL-tips | 0 | 1 | 0 | 1 |
| with <30% NL-tips | 0 | 1 | 0 | 1 |
We classified the HIV trees into trees from a skewed-clustered network with different infection rates. This has been done for a static network and for a dynamic network (δ = 0.1). We predicted the parameters for 90 HIV trees (of which 13 had 50% of tips from the Netherlands, and 7 more than 70%). Sizes of the training sets for the classifiers are 400 and 244. As in Table 6, branchlengths of simulated trees were not used. Simulated trees and anonymized HIV trees to this table are found in S6 File.