Polypilot product mascot

Introducing PolyPilot:

Our AI-Powered Mentorship Program

Start your trial today

Learn More
Go to Polygence Scholars page
Arjun Ramakrishnan's cover illustration
Polygence Scholar2022
Arjun Ramakrishnan's profile

Arjun Ramakrishnan

Academies of Loudoun/Broad Run High SchoolClass of 2023Ashburn, Virginia



  • "Hazardous Asteroid Classification with Machine Learning using Physical and Orbital Asteroid Properties" with mentor Jon (Aug. 13, 2022)

Project Portfolio

Hazardous Asteroid Classification with Machine Learning using Physical and Orbital Asteroid Properties

Started Apr. 13, 2022

Abstract or project description

Scientists are constantly studying outer space to expand their knowledge of the beginnings of the solar system. Asteroids, rocky objects orbiting the sun, have been a key focus of scientific study as they can provide insights into planet formation. With a seemingly infinite number of asteroids in space, the possibility of one colliding with our planet and leading to devastating effects constantly looms large. Asteroids that could come in close proximity or collide with earth are classified as potentially hazardous asteroids, PHA. On the most general level, this classification is determined by the size of the asteroid and by how close it actually gets to earth (NASA, n.d.). However, given the quantity of data to process when determining this classification, it becomes cumbersome for humans to manually analyze the data to make a prediction. Thus, machine learning techniques are ideal to study trends and make predictions. Machine learning is a method of data analysis based on computer algorithms that model relationships and improve our ability to analyze asteroid threats. Machine learning has been applied to automate the asteroid classification process in the past, for instance by Anish Si in 2018 at the Vellore Institute of Technology in India, where it was concluded that the 15-tree Random Forest model performed the best (Si, 2018). The application of machine learning to asteroid features is constantly evolving, emphasizing the importance of continuing to iterate on ideas that could confirm or supersede present day findings. The goal of this study was to train multiple machine learning models on asteroid features, including dimensions and orbital properties, and identify the model that most accurately classifies the asteroids as hazardous or non-hazardous. The key enhancements were that a different subset of features and significantly different list of models were used for classification. The results showed that a 50-tree Random Forest classification model had a 98.45% accuracy on the test set validating that the Random Forest is the most optimal model for asteroid classification.