ROGER: Reconstructing orbits of galaxies in extreme regions using machine learning techniques

We present the ROGER (Reconstructing Orbits of Galaxies in Extreme Regions) code, which uses three different machine learning techniques to classify galaxies in, and around, clusters, according to their projected phase-space position. We use a sample of 34 massive, M-200 > 10(15)h(-1)M(circle dot), galaxy clusters in the MultiDark Planck 2 (MDLP2) simulation at redshift zero. We select all galaxies with stellar mass M-star >= 10(8.5)h(-1)M(circle dot), as computed by the semi-analytic model of galaxy formation SAG, that are located in, and in the vicinity of, these clusters and classify them according to their orbits. We train ROGER to retrieve the original classification of the galaxies from their projected phase-space positions. For each galaxy, ROGER gives as output the probability of being a cluster galaxy, a galaxy that has recently fallen into a cluster, a backsplash galaxy, an infalling galaxy, or an interloper. We discuss the performance of the machine learning methods and potential uses of our code. Among the different methods explored, we find the K-Nearest Neighbours algorithm achieves the best performance.