Machine learning and image analysis for morphological galaxy classification

Abstract

In this paper we present an experimental study of machine learning and image analysis for performing automated morphological galaxy classification. We used a neural network, and a locally weighted regression method, and implemented homogeneous ensembles of classifiers. The ensemble of neural networks was created using the bagging ensemble method, and manipulation of input features was used to create the ensemble of locally weighed regression. The galaxies used were rotated, centred, and cropped, all in a fully automatic manner. In addition, we used principal component analysis to reduce the dimensionality of the data, and to extract relevant information in the images. Preliminary experimental results using 10-fold cross-validation show that the homogeneous ensemble of locally weighted regression produces the best results, with over 91 per cent accuracy when considering three galaxy types (E, S and Irr), and over 95 per cent accuracy for two types (E and S).