Free Access

Table 3.

Training on tabular data from Virgo and evaluating on Fornax data.

Averaged per galaxy Average over all sources

Method TPR FPR FDR AUC ROC TPR FPR FDR AUC ROC # TPs # FPs
Logistic Regression 0.63 0.05 0.33 0.93 0.70 0.05 0.16 0.94 4331 832
Support Vector Machine (linear) 0.66 0.05 0.31 0.73 0.05 0.15 4474 793
Support Vector Machine (radial) 0.73 0.04 0.23 0.81 0.04 0.12 4993 656
Nearest Neighbour 0.86 0.05 0.26 0.90 0.06 0.14 5536 915
12 Nearest Neighbours 0.86 0.04 0.19 0.98 0.91 0.04 0.11 0.98 5597 678
Decision Tree 0.87 0.05 0.24 0.97 0.89 0.05 0.13 0.98 5496 803
Random Forest 0.86 0.03 0.14 0.99 0.91 0.03 0.08 0.99 5602 470
AdaBoost 0.86 0.03 0.15 0.99 0.91 0.03 0.08 0.99 5596 473
CatBoost 0.89 0.04 0.19 0.98 0.93 0.04 0.10 0.99 5709 656
Neural Network (29-1) 0.61 0.05 0.33 0.92 0.68 0.05 0.16 0.93 4159 801
Neural Network (29-100-100-1) 0.89 0.04 0.21 0.98 0.93 0.05 0.11 0.99 5754 739

Notes. The Fornax data contains in total 21 767 sources with 6161 catalogued GCs. Results are given for a decision threshold of 0.5. Support vector machines and nearest neighbour only provide a class label and hence, no AUC ROC is reported. For neural networks, the number of neurons per layer is given in brackets.

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.