Automatic Classification of Microlensing Candidates
MetadataShow full item record
It is both exciting and important to look for life beyond our planet. To find signs of life on distant planets, there is a need to search across the vast space that surrounds us and find planets outside our solar system, called exoplanets. Among the many search techniques which have been developed to detect exoplanets, ‘microlensing’ holds the advantage of finding Earth-like planets. In order to detect a microlensing event, there is a need to scan millions of stars simultaneously for the case of perfect alignment of two stars. This chance alignment typically lasts for weeks or days, until the two stars move out of alignment. Hence, there is a need to follow up on all detected events in real-time, to capture information about the properties of the star system. Large scale astronomical surveys like the Global Astrometric Interferometer for Astrophysics (Gaia) mission and Large Synoptic Survey Telescope (LSST) will capture terabytes of data every night. Hence, building an automatic classifier, using tools from machine learning in order to sift through this data and detect microlensing events is crucial. The scope of work includes identification and development of three appropriate methods to establish an automatic classifier. The first method makes classification decisions based on five characteristics of microlensing translated into statistical features. The second and third methods detect microlensing events without relying on any specific characteristics of microlensing, but differ in the way they handle data. These methods are applied to datasets from three different astronomical surveys and the results thus obtained are evaluated to make sure that all the occurrences of this rare event, microlensing, are detected. The third method uses an RNN to detect all the events in the training set. It is concluded that, this method can be easily extended to exoplanet detection.
Timmaraju, Virisha (2018). Automatic Classification of Microlensing Candidates. Master's thesis, Texas A & M University. Available electronically from