Table 4. Prediction accuracies (correctly predicted/all predictions).
| Predicted value | Accuracy (knn) |
Accuracy (random forest) |
Size of training set | Size of test set |
|---|---|---|---|---|
| All three networks, range of mean degree and turnover | ||||
| network | 0.88 ± 0.02 | 0.92 ± 0.01 | 1084 (3 networks) | 465 (3 networks) |
| β | 0.40 ± 0.01 | 0.47 ± 0.03 | 1084 (3 networks) | 465 (3 networks) |
| Correctly specified network, range of mean degree and turnover | ||||
| β | 0.38 ± 0.04 | 0.43 ± 0.05 | 261 (skewed-clustered) | 113 (skewed-clustered) |
| β | 0.39 ± 0.04 | 0.55 ± 0.04 | 262 (skewed) | 113 (skewed) |
| β | 0.39 ± 0.03 | 0.44 ± 0.03 | 560 (random) | 240 (random) |
| Mis-specified network, range of mean degree and turnover | ||||
| β | 0.30 ± 0.01 | 0.37 ± 0.01 | 800 (random) | 374 (skewed-clustered) |
| β | 0.34 ± 0.01 | 0.39 ± 0.01 | 375 (skewed) | 374 (skewed-clustered) |
| δ | 0.36 ± 0.03 | 0.45 ± 0.03 | 1084 (3 networks) | 465 (3 networks) |
Predictions of network type, infection rate β and turnover rate δ. Values are mean and standard deviation of 10-fold cross-validation. For this, β (and δ respectively) is grouped into bins of width 0.01. β is considered to be classified correctly if it is classified into the correct or in neighbouring bins (i.e. in a range of 0.03). For the simulations, infection rate β and turnover δ are both distributed uniformly at random in [0.05, 0.15], and mean degree between [4, 9] respectively. Simulated trees to this table are found in S4 File.