When studying the classification of galaxies with unsupervised machine learning using the physical information available in data such as the SDSS MaNGA integral field spectroscopic survey, outliers appear in the results. These may result from non-physical issues or confusions in the software, but may also constitute a potent tool to find 'strange' or simply very rare galaxies.

This project is focused on different ways to detect outliers. We present the first results of this work which is part of an ongoing project that classifies MaNGA galaxies using unsupervised machine learning techniques. So far, outliers had not been studied in detail, hence when dividing the galaxy classes, outliers included in the sample could be slightly modifying the outcome. Furthermore, these outliers could be an interesting source of information to investigate, thus allowing us to find exotic galaxies. Therefore, developing a method to detect outliers is crucial both to improving the classification and the study of unique galaxies. We use different methods for anomaly detection and present preliminary results.