SUPERVISED DETECTION OF ANOMALOUS LIGHT CURVES IN MASSIVE ASTRONOMICAL CATALOGS

Abstract

The development of synoptic sky surveys has led to a massive amount of data\nfor which resources needed for analysis are beyond human capabilities. To\nprocess this information and to extract all possible knowledge, machine\nlearning techniques become necessary. Here we present a new method to\nautomatically discover unknown variable objects in large astronomical catalogs.\nWith the aim of taking full advantage of all the information we have about\nknown objects, our method is based on a supervised algorithm. In particular, we\ntrain a random forest classifier using known variability classes of objects and\nobtain votes for each of the objects in the training set. We then model this\nvoting distribution with a Bayesian network and obtain the joint voting\ndistribution among the training objects. Consequently, an unknown object is\nconsidered as an outlier insofar it has a low joint probability. Our method is\nsuitable for exploring massive datasets given that the training process is\nperformed offline. We tested our algorithm on 20 millions light-curves from the\nMACHO catalog and generated a list of anomalous candidates. We divided the\ncandidates into two main classes of outliers: artifacts and intrinsic outliers.\nArtifacts were principally due to air mass variation, seasonal variation, bad\ncalibration or instrumental errors and were consequently removed from our\noutlier list and added to the training set. After retraining, we selected about\n4000 objects, which we passed to a post analysis stage by perfoming a\ncross-match with all publicly available catalogs. Within these candidates we\nidentified certain known but rare objects such as eclipsing Cepheids, blue\nvariables, cataclysmic variables and X-ray sources. For some outliers there\nwere no additional information. Among them we identified three unknown\nvariability types and few individual outliers that will be followed up for a\ndeeper analysis.\n