The Wide Field Infrared Survey Telescope (WFIRST) is a NASA flagship mission scheduled to launch in mid-2020, with more than one year of its lifetime dedicated to microlensing survey. The aim is to discover thousands of exoplanets via their microlensing lightcurves, which will enable a Kepler-like statistical analysis of planets ~1-10 AU from their parent stars and revolutionize theories of planet formation. The goal of our work is to create an automated system that has the ability to efficiently process and classify large-scale astronomical datasets that missions such as WFIRST will produce. In this paper, we discuss our framework that utilizes feature selection and parameter optimization for classification models to automatically differentiate the different types of stellar variability and detect microlensing events. The use of feature selection enables us to learn which characteristics distinguish the different types of events and to classify high-dimensional data more efficiently. We demonstrate our proposed method on datasets acquired from UKIRT’s wide-field near-IR camera that surveys the galactic bulge.
Link to PDF (may not be available yet): P1-21.pdf