Prediction of Orbital Parameters for Undiscovered Potentially Hazardous Asteroids Using Machine Learning

Abstract

The purpose of this study is to make a prediction of combinations of orbital parameters for yet undiscovered potentially hazardous asteroids (PHAs) with the use of machine learning algorithms. The proposed approach aims at outlining subgroups of all major groups of near-Earth asteroids (NEAs) with high concentration of PHAs in them. The approach is designed to obtain meaningful results and easy-understandable boundaries of the PHA subgroups in 2- and 3-dimensional subspaces of orbital parameters. Boundaries of these PHA subgroups were found mainly by the use of Support Vector Machines algorithm with RBF kernel. Additional datasets of virtual asteroids were generated to handle sufficient amount of training and test data, as well as to emulate undiscovered asteroids. This synthetic data helped in revealing ‘XX’-shaped region with high concentration of PHAs in the (ω, q) plane. Boundaries of this region were used to split all NEAs into several domains. For each domain the subgroups of PHAs were outlined in different subspaces of orbital parameters. Extracted subgroups have high PHA purity (∼90%) and contain ∼90% of all real and virtual PHAs. Obtained results can be useful for planning future PHA discovery surveys or asteroid-hunting space missions.