Context. The huge and still rapidly growing amount of galaxies in modern sky surveys raises the need for an automated and objective classification method. Unsupervised learning algorithms are of particular interest, since they discover classes automatically. Aims. We briefly discuss the pitfalls of oversimplified classification methods and outline an alternative approach called "clustering analysis". Methods. We have categorised different classification methods according to their capabilities. Based on this categorisation, we present a probabilistic classification algorithm that automatically detects the optimal classes preferred by the data. We explored the reliability of this algorithm in systematic tests. Using a sample of 1520 bright galaxies from the SDSS, we demonstrate the performance of this algorithm in practice. We are able to disentangle the problems of classification and parametrisation of galaxy morphologies in this case. Results. We give physical arguments that a probabilistic classification scheme is necessary. When applied to a small set of 84 galaxies visually classified as face-on discs, edge-on discs, and ellipticals, the clustering algorithm discovers precisely these classes and produces excellent object-to-class assignments. The resulting grouping of the galaxies outperforms a principal components analysis applied to the same data set. Applying the algorithm to a sample of 1520 SDSS galaxies, we find morphologically distinct classes when the number of classes are 3 and 8. Conclusions. Although interpreting clustering results is a challenging task, the algorithm we present produces reasonable morphological classes and object-to-class assignments without any prior assumptions.
All Science Journal Classification (ASJC) codes
- Astronomy and Astrophysics
- Space and Planetary Science
- methods: data analysis
- methods: statistical