Statistical analyses of the evolution of galactic appearance (or morphology) play an important role in cosmology in many circumstances, e.g. when trying to infer cosmological parameters. Automatic detection of some morphologies of galaxies can be done efficiently in the nearby universe using previously developed image-based summary statistics (see, e.g., Conselice, 2003 and Lotz et al., 2004). However, to perform precise statistical inference, we must be able to classify galaxies at much greater distances, and the ability to detect different morphologies using current image-based summary statistics decreases with distance. The aim of this work is to develop new summary statistics and build classifiers that allow one to distinguish galaxies that are irregulars, mergers or interactions from other types even in the farther universe. First, we propose new more informative features. Then, we apply 4 different classification algorithms to the data set. We show that the results for all of them are consistently better than when using only traditional features. Finally, we show that the classifications can be further improved if the classifier is allowed to exclude a fraction of the data set which is specified by the user. In practice, this subset could be classified by experts. We also evaluate how the new features perform when trying to classify mergers/interactions only.