Table D.5.
Comparison between our image-based CNN model and two different photometric catalogue-based approaches, referred to the EXP3 experiment.
| R2248 z = 0.346 | |||||
|---|---|---|---|---|---|
| CNN | RF | Bayesian | Δ | ||
| AE | 88.1 | 86.5 | 85.9 | 1.6 | |
| pur | 88.3 | 87.7 | 80.9 | 0.6 | |
| CLM | compl | 89.8 | 87.7 | 96.1 | −6.3 |
| F1 | 89.1 | 87.7 | 87.8 | 1.3 | |
| pur | 87.9 | 85.1 | 94.4 | −6.5 | |
| NCLM | compl | 86.1 | 85.1 | 74.4 | 1.0 |
| F1 | 87.0 | 85.1 | 83.2 | 1.9 | |
| μΔ | −0.91 ± 1.42 | ||||
| M0416 z = 0.397 | |||||
| CNN | RF | Bayesian | Δ | ||
| AE | 92.2 | 89.2 | 87.1 | 3.0 | |
| pur | 93.3 | 93.0 | 84.6 | 0.3 | |
| CLM | compl | 87.1 | 86.5 | 91.2 | −4.1 |
| F1 | 91.5 | 89.7 | 87.8 | 1.8 | |
| pur | 89.0 | 84.5 | 90.0 | −1.0 | |
| NCLM | compl | 96.9 | 92.3 | 82.7 | 4.6 |
| F1 | 91.5 | 88.3 | 86.2 | 3.2 | |
| μΔ | 1.11 ± 1.12 | ||||
| M1206 z = 0.439 | |||||
| CNN | RF | Bayesian | Δ | ||
| AE | 89.7 | 87.9 | 85.0 | 1.8 | |
| pur | 89.9 | 90.4 | 80.2 | −0.5 | |
| CLM | compl | 86.5 | 81.9 | 91.2 | −4.7 |
| F1 | 88.2 | 85.9 | 85.3 | 2.3 | |
| pur | 89.6 | 86.3 | 90.8 | −1.2 | |
| NCLM | compl | 92.3 | 92.9 | 79.4 | −0.6 |
| F1 | 90.9 | 89.7 | 84.7 | 1.2 | |
| μΔ | −0.24 ± 0.90 | ||||
| M1149 z = 0.542 | |||||
| CNN | RF | Bayesian | Δ | ||
| AE | 89.4 | 86.9 | 85.5 | 2.5 | |
| pur | 82.3 | 78.8 | 71.8 | 3.5 | |
| CLM | compl | 91.3 | 88.5 | 98.0 | −6.7 |
| F1 | 86.6 | 83.4 | 82.9 | 3.2 | |
| pur | 94.5 | 92.7 | 98.6 | −4.1 | |
| NCLM | compl | 88.3 | 86.0 | 78.4 | 2.3 |
| F1 | 91.3 | 83.4 | 87.4 | 3.9 | |
| μΔ | 0.66 ± 1.60 | ||||
Notes. The comparison involves two different model: a Random Forest and a Bayesian method, applied on photometric tabular information of four clusters: R2248 (z = 0.346), M0416 (z = 0.397), M1206 (z = 0.439) and M1149 (z = 0.542). Last column (Δ) shows the difference between CNN estimators and the best between the two photometric approaches, i.e. Δestim = estimCNN − max{estimRF, estimBayesian} for estim ∈ [pur, compl, F1, AE], while rows μΔ list the averages among these Δs for each cluster.
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.