Abstract |
Classification methods for characterizing the activity of a galaxy are of high importance in observational astrophysics. Even though numerous diagnostic tools have been build over the past years, the overwhelming majority of them only concerns emission line galaxies or are specialized only in
one activity class (i.e, AGN). Moreover, almost none of them is able to include all possible types of activity (active and passive) under one unified scheme. Furthermore, they fail to properly address the issue of the mixed activity classes of composite and LINER galaxies. In this work, we intent to define a diagnostic tool based on machine-learning methods considering three classes that are representative of the principal mechanisms of gas excitation: star-formation, active nucleus and excitation from hot
evolved stars present in passive galaxies. We use data from the SDSS and GALEX All-sky surveys in order to select the training sample of the active and passive galaxies. For this purpose, we train a Random Forest algorithm that utilizes four features in total. Three of them are the Equivalent
Widths (EW) of the spectral lines of Hα, [NII] λ6584Å, and [OIII] λ5007Å that are found to provide excellent discriminating power for the three principal types of the activity found in a galaxy. The fourth feature is the D4000 continuum break index which is a good indicator of the average age of the stellar populations. We manage to achieve accuracy of ∼ 99%. Due to the high performance scores achieved on the pure activity classes and based on the predicted probabilities provided by the Random
Forest we can apply this method to the mixed activity classes in order to identify the dominant source of gas excitation in a galaxy. For this reason we also increase the considered activity classes to provide refined predictions for the mixed activity classes. Therefore, besides the bona-fide activity classes of star-forming (SF), active nucleus (AGN) and passive galaxies, we add mixed activity classes that are descriptive not only about the dominant but also for the secondary excitation mechanism that
manages to provide considerable contribution to the resultant galaxy spectrum. Finally, we apply our diagnostic tool on a sample of spectroscopically selected composite and LINER galaxies to verify the
validity of the activity decomposition of these mixed activity galaxy classes. We find that the activity decomposition of galaxies is actually possible.
|