3/8/2021

Microlensing is a powerful tool for discovering cold exoplanets, and the Roman microlensing survey will discover over 1000 such planets. The full modeling of each planetary microlensing event often requires significant investment of human and computing resources. When Roman releases light curves of thousands of microlensing events, it is important to detect the planetary systems fast, so an algorithm is needed to quickly classify all microlensing signals and prioritize rapid follow-up observations.  Machine learning (ML) is becoming more popular in astronomy for classification purposes. Using some of these ML algorithms usually require having meaningful features defined for the objects that are to be classified. In this presentation, I will talk about algorithms we have developed to capture features in simulated Roman light curves of different types of microlensing events and evaluate their effectiveness. These features are quantified as parameters that can be used to decide how likely a light curve contains a specific type of microlensing event. This method leaves us with a list of parameters that describe features like smoothness of the peak, symmetry, number of peaks, and width and height of the small deviations from the main peak. This will allow us to quickly analyze a set of microlensing light curves and later use the resulting parameters as input to machine learning algorithms to classify the events.