Jairam Susarla
7th Grade, The Honor Roll School
Sugar Land, TX
Jairam has been an astronomy fan for most of his life and was thrilled to learn about exoplanets — planets outside of our solar system — and “how several thousand had been discovered in just the past few years,” he says. Most are discovered by space telescopes on the hunt for transits — the dark shadow an exoplanet will cast as it moves in front of its star. “Although the Transit Method is very accurate, the time it takes to validate an exoplanet from telescope data can take several months,” Jairam says. He decided to use machine learning to find a faster way.
Comparing ML Models to Uncover Exoplanets
View PosterProject Background
Jairam used data from the Kepler Space Telescope Exoplanet Hunting in Deep Space dataset. The dataset has measures of light changes from 5,087 stars, all of which have already been classified with whether or not they have an exoplanet. He divided up the data and used 90 percent of it to train six different machine learning models. Then he used the last 10 percent of the data to test the models once they were trained.
Jairam showed that one of the models, the convoluted neural network (CNN), had 99.82 percent accuracy. Its precision score was 0.83 (meaning that it said one out of six stars had an exoplanet when they did not). He hopes that his model will eventually make detecting and validating new exoplanets easier and much faster.
Beyond the Project
Jairam is a brown belt in Kuk Sool Won, a Korean martial art. He loves buzzer-based science competitions like the National Science Bee, where he always picks astronomy as his topic. He was deeply inspired by Carl Sagan’s “Cosmos” series and hopes to become an astrophysicist one day. “The possibility of finding planets outside of our solar system and life on other planets is very exciting to me,” he says.