Table 2.
Comparison of two feature sets: all 83 features and a limited set of 47 features excluding magnitude ratios, reported on the random (r < 21.3) and faint extrapolation (r ∈ (21.3, 22)) tests.
| Classification |
Redshift for true QSOs |
Redshift for QSO candidates |
|||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Test | Model | Features | Accuracy | Purity | Recall | MSE | R2 | δz | MSE | R2 | δz |
| Random | RF | All features | 99.01% | 97.39% | 94.43% | 0.12 | 85% | 0.02 ± 0.14 | 0.12 | 84% | 0.03 ± 0.21 |
| No ratios | 98.98% | 97.20% | 94.57% | 0.11 | 86% | 0.02 ± 0.14 | 0.14 | 82% | 0.04 ± 0.25 | ||
| XGB | All features | 99.09% | 97.80% | 95.82% | 0.13 | 84% | 0.02 ± 0.15 | 0.13 | 83% | 0.03 ± 0.21 | |
| No ratios | 99.07% | 97.80% | 94.66% | 0.12 | 84% | 0.01 ± 0.15 | 0.12 | 83% | 0.03 ± 0.20 | ||
| ANN( * ) | All features( * ) | 98.98% | 96.90% | 94.72% | 0.09 | 88% | 0.01 ± 0.12 | 0.11 | 85% | 0.02 ± 0.23 | |
| No ratios | 98.99% | 97.58% | 94.30% | 0.09 | 88% | 0.01 ± 0.12 | 0.11 | 86% | 0.02 ± 0.21 | ||
| Faint | RF | All features | 97.41% | 96.07% | 92.30% | 0.31 | 31% | 0.02 ± 0.25 | 0.33 | 31% | 0.05 ± 0.38 |
| extrap. | No ratios | 97.39% | 96.12% | 92.22% | 0.29 | 35% | 0.01 ± 0.24 | 0.31 | 35% | 0.04 ± 0.38 | |
| XGB | All features | 97.41% | 96.44% | 92.07% | 0.27 | 39% | 0.04 ± 0.23 | 0.34 | 29% | 0.08 ± 0.41 | |
| No ratios | 97.45% | 96.55% | 91.94% | 0.27 | 39% | 0.04 ± 0.22 | 0.33 | 29% | 0.08 ± 0.40 | ||
| ANN( * ) | All features( * ) | 97.24% | 96.54% | 90.82% | 0.22 | 51% | 0.00 ± 0.19 | 0.28 | 38% | 0.04 ± 0.37 | |
| No ratios | 97.26% | 96.94% | 90.88% | 0.24 | 46% | 0.00 ± 0.20 | 0.30 | 32% | 0.03 ± 0.38 | ||
Notes. The redshifts were tested in two subsets of QSOs: true spectroscopic ones and photometric candidates. The candidates include misclassified sources, e.g., a true star assigned a QSO class and redshift. We mark with bold font the best results, independently for random and faint extrapolation tests. The asterisk (*) marks the approach used to create the final catalog. Recall is the same as completeness, and MSE, R2, and δz are given by Eqs. (1)–(3), respectively.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.