http://arxiv.org/abs/1712.08955

A galaxy morphological type is correlated with the color indices, luminosity, de Vaucouleurs radius, inverse concentration index etc. To study these relations we have to operate with big samples of galaxies, so the visual morphological inspection is not always possible. We evaluated a new approach. Namely, we applied the “color–concentration index” diagram and machine learning methods for the morphological classification of galaxies from the SDSS at z<0.1. With this aim, we visually identified morphological T-types of about 1500 galaxies, which formed our training samples. Method 1. We plotted the diagrams of color indices g-i and one of such parameters as the inverse concentration index, absolute magnitude, de Vaucouleurs radius. We discovered that these parameters may be used for galaxy classification into three classes: E — elliptical and lenticular, S — types Sa-Scd, and L — types Sd-Sdm and irregular s. The accuracy is 98% for E, 88% for S, and 57% for L types. The combinations of “color indices g-i and inverse concentration index R50/R90′ and “color indices g-i and absolute magnitude M_r” give the best result. We applied this method to classify 317018 galaxies from SDSS DR5 (143263 E, 112 578 S, 61177 L types). Method 2. We used a training sample classified visually into two classes: early E (E, S0, S0a) and late L (Sa to Irr) types. We checked Naive Bayes, Random Forest, and Support Vector Classifier. We used absolute magnitudes, all the color indices and inverse concentration indexes as the attributes of galaxy. To define an accuracy of classifiers we applied the 5-folds validation and found that Random Forest provides the highest accuracy (91% of galaxies were correctly classified (96% for E and 80% for L types)). We tested it to classify 60561 galaxies from SDSS DR9 with a good accuracy onto two classes (47% E and 53% L types of galaxies).

Read this paper on arXiv…

D. Dobrycheva, I. Vavilova, O. Melnyk, et. al.
Wed, 27 Dec 2017
49/56

Comments: 4 pages, 5 figures. The presentation of these results was given during the EWASS-2017, Symposium “Astroinformatics: From Big Data to Understanding the Universe at Large”. It is vailable through \url{this http URL}