NEOTracker

Abstract

Near-Earth objects (NEOs) are asteroids or comets that have their orbits in close proximity with Earth. Some objects amongst these are known to be potentially hazardous and pose a risk of collision. This chapter developed four supervised machine learning algorithms, namely, logistic regression, random forest, support vector machine, and XGBoost, for the detection and classification of hazardous near-earth objects. Two datasets were utilised, the first taken from the Kaggle website, and the second generated from NASA's JPL Small-Body database. Feature importance analysis of these datasets was done by analysing the Shapley values of the individual features in both datasets. This chapter concludes by finding all models to have performed sufficiently well, with XGBoost found to be the best and most consistent performing across both datasets. Additionally, both min and max diameter, and the absolute magnitude features for the Kaggle dataset, and the H and moid features for the JPL dataset were found to be the most impactful features for classifying hazardous near-earth objects.