Lukas Wenzl and colleagues presented a a supervised machine learning method, called “random forests” for selection and discovery of high redshift quasars. They argued that random forests can lead to higher completeness and shows promise in finding quasars that would be missed by traditional approaches like color cuts. Their study recently appeared in Arxiv.
Many selection methods for high redshift quasars make use of the broadband colors and magnitudes of large photometric catalogs and combine them with information about the morphology, time variability, X-ray or radio detections, position and proper motion. Sophisticated color cuts define selection regions in color-color space to separate quasar and contaminant distributions. This leads to well-defined selections that are easily reproducible and can be justified with physical reasoning. However, color cuts might not make use of all the available information, by ignoring correlations in the full high dimensional color space. Furthermore, they represent hard cuts, potentially missing quasars, scattering out of the selection regions.
Now, Wenzl and colleagues presented a method of selecting quasars up to redshift ≈ 6 with random forests, a supervised machine learning method, applied to Pan-STARRS1 and WISE data. They demonstrated that, the method can enable the discovery of more quasars.
“We choose random forests for their robustness and fast training”
In addition, it shows promise in finding quasars that would be missed by traditional approaches like color cuts. One of their newly discovered z=5.7 quasars (J152330.66+293539.1) would be rejected by a common cut on the i-z color for z > 5.6 quasars.
“Our high-redshift candidate set contains more promising candidates that would be rejected by that cut. Therefore our random forest approach shows promise to reach higher completeness and is relevant for future quasar luminosity estimates”
Moreover, spectroscopic follow-up observations of 37 candidates lead them to discover 20 new high redshift quasars (18 at 4.6 ≤ z ≤ 5.5, 2 z ∼ 5.7). These observations are consistent with their predictions on efficiency.
(article continues below image)
Finally, they argued that random forests can lead to higher completeness because their candidate set contains a number of objects (about 148 new quasars below redshift 5.6 and 45 above) that would be rejected by common color cuts, including one of the newly discovered redshift 5.7 quasars.
The python code for this project is available under https://github.com/lukaswenzl/High-Redshift-Quasars-with-Random-Forests
Reference: Lukas Wenzl, Jan-Torge Schindler, Xiaohui Fan, Irham Taufik Andika, Eduardo Banados, Roberto Decarli, Knud Jahnke, Chiara Mazzucchelli, Masafusa Onoue, Bram P. Venemans, Fabian Walter, Jinyi Yang, “Random Forests as a viable method to select and discover high redshift quasars”, Arxiv, pp. 1-25, 2021. https://arxiv.org/abs/2105.09171
Note for editors of other websites: To reuse this article fully or partially kindly give credit either to our author/editor S. Aman or provide a link of our article