A Data-driven Approach to the Classification of Temporary Captures in the Earth-Moon System

This paper outlines a new classification approach to characterizing the temporary capture of asteroids in the Earth-Moon system based on the simulation data. Through the use of machine learning classification techniques, feature selection shows that synodic state vectors perform better than heliocentric state vectors for various classification tasks. The optimized classifier, which combines the random forest and histogram gradient boosting algorithms, enables the prediction of long-lived and short-lived temporary captures as well as the prediction of temporary captures that will come closer than the Moon's orbit to Earth.