Spectral classification indicators of emission-line galaxies from the Sloan Digital Sky Survey

To find efficient spectral classification diagrams to classify emission-line galaxies, especially in large surveys and huge data bases, an artificial neural network (ANN) supervised learning algorithms is applied to a sample of emission-line galaxies from the Sloan Digital Sky Survey data release 9 provided by the Max Planck Institute and the Johns Hopkins University (MPA/JHU) (http://www.sdss3.org/dr9/spectro/spectroaccess.php). A two-step approach is adopted. (i) The ANN network must be trained with a subset of objects that are known to be active galactic nuclei (AGNs) hosts, composites or star-forming galaxies, treating the strong emission-line flux measurements as input feature vectors in n-dimensional space, where n is the number of strong emission-line flux ratios. (ii) After the network is trained on a sample of galaxies, the remaining galaxies are classified in the automatic test analysis as AGN hosts, composites or star-forming galaxies. We show that the classification diagrams based on the [N ii]/Hα versus other emission-line ratio, such as [O iii]/Hβ, [Ne iii]/[O ii], ([O iii]λ4959 + [O iii]λ5007)/[O iii]λ4363, [O ii]/Hβ, [Ar iii]/[O iii], [S ii]/Hα, and [O i]/Hα, plus colour, allows us to separate unambiguously AGN hosts, composites or star-forming galaxies. Among them, the diagram of [N ii]/Hα versus [O iii]/Hβ achieved an accuracy of 98 per cent for classification of AGN hosts, composites or star-forming galaxies. The other diagrams above except the diagram of [N ii]/Hα versus [O iii]/Hβ give an accuracy of ∼90 per cent. The code in the paper is available on the web (http://fshi5388.blog.163.com).