Asteroid Orbit Classification with Machine Learning: A Data-Driven Approach

Classifying asteroid orbits is crucial for understanding solar system dynamics and improving planetary defense mechanisms. Traditional classification methods, which depend on manual calculations, struggle to keep up with the growing volume and complexity of asteroid data. This paper tackles these challenges by using machine learning to classify asteroid orbits based on the dataset from the Minor Planet Center, which is marked by significant class imbalance and high dimensionality. The preprocessing phase involved thorough data cleaning, addressing missing values and eliminating irrelevant features, which resulted in a reduction of the dataset from 34 to 12 features. Furthermore, to address the imbalance, majority classes were under-sampled, while minority classes like Atira were over-sampled. Five supervised learning models-Decision Trees, K-Nearest Neighbors (KNN), Multi-layer Perceptron (MLP), Naïve Bayes, and Random Forest-were assessed. The MLP model achieved 100% overall accuracy, surpassing the performance of the other models, whereas KNN fell short at 83% due to its sensitivity to class imbalance. This work illustrates the effectiveness of advanced preprocessing and MLP in overcoming dataset limitations, providing a scalable solution for accurate asteroid classification.