Classification of Astronomical Objects Using Various Machine Learning Techniques

Abstract

In recent times, enormous chunks of data is being extracted from the astronomical telescopes throughout the world for better and more precise information on the astronomical object so as to classify the object. As advances in data science have been rapid, the field of data mining in astronomy (Borne in Scientific data mining in astronomy. George Mason University Fairfax, USA, 2009 [1]) has also been on an upward slope. Papers on data mining in astronomy and classification of astronomical objects (Mahabal et al. in towards real-time classification of astronomical transients, 2008 [2]; D’Isanto et al. in An analysis of feature relevance in classification of astronomical transients with machine learning methods, 2016 [3]) have paved a way for smarter manipulation of available data to extract better results. In the previous years, Artificial Neural Networks and Support Vector Machines (SVM) have been used on astronomical data to handle classification problems in the available data whereas in some cases imaging technology and lightcurves (Faraway et al. in Modern light curves for improved classification, 2014 [4]) have been used for the given problem. Image Processing has also been used for the morphological classification of galaxy images (Shamir in Automatic morphological classification of galaxy images, 2009 [5]). Our motivation is to find the which machine learning algorithm uses the least computation time and gives the best accuracy metric for classification of astronomical objects into Stars, Galaxies and Quasars based on data provided to us by the Sloan Digital Sky Survey/SkyServer (SDSS). This research gives us the best suited classification algorithm out of all the present and currently used algorithms. We have used Logistic Regression, Support Vector Machines, Random Forests and Decision Tree classifiers and compared the results obtained to come to the conclusion of the most suitable classification algorithm for the given astronomical data.