A Hierarchical Approach for Multi-class Galaxy Classification

As the universe continues to evolve, galaxies are constantly being created and destroyed. This calls for the need for a system that automatically mimics experts and categorizes galaxies. In this paper we propose a hierarchical approach to classification of multiple classes of galaxies. Our method combines features from principal component analysis as well as domain knowledge. Additionally, we use single-class classification to examine the accuracy of class-labels generated by a decision tree, and we examine several classifiers to determine that classifier which best suits our purpose. The overall approach leads to reasonably good accuracies in our context, and to the best of our knowledge, better than recorded elsewhere.