Cross-identification accuracies for different classification models on ELAIS-S1. Columns and abbreviations are as in Table B3. Accuracies are evaluated against the expert label set derived from Middelberg et al. (2008) cross-identifications. The standard deviation of accuracies evaluated across models trained on the four quadrants of CDFS (Fig. 8) is also shown.
| Labeller . | Classifier . | Mean ‘compact’ accuracy . | Mean ‘resolved’ accuracy . | Mean ‘all’ accuracy . |
|---|---|---|---|---|
| . | . | (per cent) . | (per cent) . | (per cent) . |
| – | NN | 95.5 ± 0.0 | 92.8 ± 0.0 | 95.5 ± 0.0 |
| – | Random | 61.9 ± 1.1 | 26.6 ± 2.1 | 61.9 ± 1.1 |
| Middelberg | Perfect | 99.6 ± 0.0 | 99.8 ± 0.0 | 99.6 ± 0.0 |
| Norris | LR | 89.0 ± 1.1 | 89.7 ± 1.8 | 94.4 ± 0.9 |
| CNN | 89.7 ± 0.3 | 89.4 ± 1.4 | 94.3 ± 0.7 | |
| RF | 83.8 ± 5.6 | 82.3 ± 4.1 | 90.6 ± 2.1 | |
| RGZ | LR | 90.5 ± 1.0 | 92.7 ± 0.2 | 95.9 ± 0.1 |
| CNN | 84.6 ± 0.6 | 84.6 ± 0.6 | 91.8 ± 0.3 | |
| RF | 91.3 ± 1.0 | 90.3 ± 2.4 | 94.7 ± 1.2 |
| Labeller . | Classifier . | Mean ‘compact’ accuracy . | Mean ‘resolved’ accuracy . | Mean ‘all’ accuracy . |
|---|---|---|---|---|
| . | . | (per cent) . | (per cent) . | (per cent) . |
| – | NN | 95.5 ± 0.0 | 92.8 ± 0.0 | 95.5 ± 0.0 |
| – | Random | 61.9 ± 1.1 | 26.6 ± 2.1 | 61.9 ± 1.1 |
| Middelberg | Perfect | 99.6 ± 0.0 | 99.8 ± 0.0 | 99.6 ± 0.0 |
| Norris | LR | 89.0 ± 1.1 | 89.7 ± 1.8 | 94.4 ± 0.9 |
| CNN | 89.7 ± 0.3 | 89.4 ± 1.4 | 94.3 ± 0.7 | |
| RF | 83.8 ± 5.6 | 82.3 ± 4.1 | 90.6 ± 2.1 | |
| RGZ | LR | 90.5 ± 1.0 | 92.7 ± 0.2 | 95.9 ± 0.1 |
| CNN | 84.6 ± 0.6 | 84.6 ± 0.6 | 91.8 ± 0.3 | |
| RF | 91.3 ± 1.0 | 90.3 ± 2.4 | 94.7 ± 1.2 |
Cross-identification accuracies for different classification models on ELAIS-S1. Columns and abbreviations are as in Table B3. Accuracies are evaluated against the expert label set derived from Middelberg et al. (2008) cross-identifications. The standard deviation of accuracies evaluated across models trained on the four quadrants of CDFS (Fig. 8) is also shown.
| Labeller . | Classifier . | Mean ‘compact’ accuracy . | Mean ‘resolved’ accuracy . | Mean ‘all’ accuracy . |
|---|---|---|---|---|
| . | . | (per cent) . | (per cent) . | (per cent) . |
| – | NN | 95.5 ± 0.0 | 92.8 ± 0.0 | 95.5 ± 0.0 |
| – | Random | 61.9 ± 1.1 | 26.6 ± 2.1 | 61.9 ± 1.1 |
| Middelberg | Perfect | 99.6 ± 0.0 | 99.8 ± 0.0 | 99.6 ± 0.0 |
| Norris | LR | 89.0 ± 1.1 | 89.7 ± 1.8 | 94.4 ± 0.9 |
| CNN | 89.7 ± 0.3 | 89.4 ± 1.4 | 94.3 ± 0.7 | |
| RF | 83.8 ± 5.6 | 82.3 ± 4.1 | 90.6 ± 2.1 | |
| RGZ | LR | 90.5 ± 1.0 | 92.7 ± 0.2 | 95.9 ± 0.1 |
| CNN | 84.6 ± 0.6 | 84.6 ± 0.6 | 91.8 ± 0.3 | |
| RF | 91.3 ± 1.0 | 90.3 ± 2.4 | 94.7 ± 1.2 |
| Labeller . | Classifier . | Mean ‘compact’ accuracy . | Mean ‘resolved’ accuracy . | Mean ‘all’ accuracy . |
|---|---|---|---|---|
| . | . | (per cent) . | (per cent) . | (per cent) . |
| – | NN | 95.5 ± 0.0 | 92.8 ± 0.0 | 95.5 ± 0.0 |
| – | Random | 61.9 ± 1.1 | 26.6 ± 2.1 | 61.9 ± 1.1 |
| Middelberg | Perfect | 99.6 ± 0.0 | 99.8 ± 0.0 | 99.6 ± 0.0 |
| Norris | LR | 89.0 ± 1.1 | 89.7 ± 1.8 | 94.4 ± 0.9 |
| CNN | 89.7 ± 0.3 | 89.4 ± 1.4 | 94.3 ± 0.7 | |
| RF | 83.8 ± 5.6 | 82.3 ± 4.1 | 90.6 ± 2.1 | |
| RGZ | LR | 90.5 ± 1.0 | 92.7 ± 0.2 | 95.9 ± 0.1 |
| CNN | 84.6 ± 0.6 | 84.6 ± 0.6 | 91.8 ± 0.3 | |
| RF | 91.3 ± 1.0 | 90.3 ± 2.4 | 94.7 ± 1.2 |
This PDF is available to Subscribers Only
View Article Abstract & Purchase OptionsFor full access to this pdf, sign in to an existing account, or purchase an annual subscription.
