Classifying High-cadence Microlensing Light Curves I; Defining Features

TL;DR

This paper develops feature extraction tools for simulated Roman Space Telescope microlensing light curves, enabling rapid classification of events to prioritize follow-up observations using machine learning.

Contribution

It introduces a set of parameters capturing key features of microlensing light curves, facilitating automated classification of different event types.

Findings

Features effectively distinguish microlensing event types

Parameters include peak smoothness, symmetry, and deviations

Method supports quick analysis for machine learning classification

Abstract

Microlensing is a powerful tool for discovering cold exoplanets, and the The Roman Space Telescope microlensing survey will discover over 1000 such planets. Rapid, automated classification of Roman's microlensing events can be used to prioritize follow-up observations of the most interesting events. Machine learning is now often used for classification problems in astronomy, but the success of such algorithms can rely on the definition of appropriate features that capture essential elements of the observations that can map to parameters of interest. In this paper, we introduce tools that we have developed to capture features in simulated Roman light curves of different types of microlensing events, and evaluate their effectiveness in classifying microlensing light curves. These features are quantified as parameters that can be used to decide the likelihood that a given light curve is due to a specific type of microlensing event. This method leaves us with a list of parameters that describe features like the smoothness of the peak, symmetry, the number of peaks, and width and height of 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.