Machine Learning Classification of Kuiper Belt Populations.

Area of Science:

• Planetary Science
• Astronomy
• Computational Astrophysics

Background:

• The Kuiper Belt harbors diverse dynamical sub-populations shaped by planetary evolution and gravitational influences.
• Classifying Kuiper Belt Objects (KBOs) traditionally relies on extensive numerical orbital integrations.

Purpose of the Study:

• To investigate the efficacy of machine learning algorithms for classifying KBOs.
• To reduce the computational and human resources required for KBO dynamical classification.

Main Methods:

• Utilized a Gradient Boosting Classifier trained on features from short numerical simulations.
• Classified 542 securely identified KBOs into four broad dynamical populations: classical, resonant, detached, and scattering.
• Analyzed object clones to derive class membership distributions, accounting for observational errors.

Main Results:

• Achieved >97% accuracy in classifying KBOs into four distinct dynamical populations.
• Over 80% of classified objects showed >3 sigma probability of class membership, indicating robust classification.
• Identified orbital ambiguity and training set limitations as primary sources of misclassification.

Conclusions:

• Machine learning offers a fast and accurate method for classifying KBOs.
• This technique is well-suited for handling the increasing number of KBO discoveries.
• The developed method can efficiently analyze large datasets and inform our understanding of Kuiper Belt dynamics.