Exploring Asteroid Orbits: Insights from Neural Network Modeling and Data-driven Analysis

In this study, orbital data from the NASA Jet Propulsion Lab was used in the classification of asteroid orbits through a machine-learning approach.

Classifying the orbits of asteroids contributes to research on the Solar System’s formation and evolution, making the classification of orbits a fundamental aspect of space exploration. However, accurate orbit determination is often hindered by skewed observational data stemming from uneven and limited sky coverage. In this study, orbital data from the NASA Jet Propulsion Lab was used in the classification of asteroid orbits through a machine-learning approach. Due to imbalances in the dataset, Synthetic Minority Oversampling Technique (SMOTE) was used to compensate for limited observations of certain orbit types. Afterwards, several classification models were implemented using specific orbital features. The most accurate results in this study were produced by a custom Keras neural network, with similar results demonstrated by an MLP neural network and k-nearest neighbors model. The custom neural network was able to effectively distinguish between orbital patterns, as demonstrated by its 99.18% validation accuracy and nearly diagonal confusion matrix. The capability of these models not only contributes to our understanding of asteroid orbits but also suggests potential improvements in orbit determination methodologies.

Explore More!

Research Paper