How did Yashasvi Moon, a senior physics student at Missouri State University, identify 9,548 galaxies without looking into a single telescope? 

She trained machine learning (ML) models to identify specific galaxies on their own. 

The American Astronomical Society recently published her research on finding quiescent Balmer-strong (QBS) galaxies. 

Combining three majors  

With a last name like Moon, it is easy to assume she would want to be an astronomer. In reality, Moon’s first interest was math.  

When Moon moved from Nagpur, India, to Springfield in January 2021, she started with a major in physics. She soon majored in math as well. As she created astronomy projects using her physics knowledge, she added astronomy as a third major.  

 Moon recalls finding an interest in physics as a young student.  

“I remember making a barometer in middle school,” she said. “It helped me take measurements of abstract quantities and turn them into something concrete.”  

In summer 2024, Moon applied for the Illinois Developing Equity in Astrophysics Summer school, where she developed her research. The program gave her a grant through the National Space Grant College and Fellowship Project. NASA funds this project. 

The summer program also connected Moon with a mentor, Dr. Decker French, from the University of Illinois Urbana-Champaign. French has done extensive research on total disruption events (TDEs) and post-starburst galaxies. Moon built upon French’s ML model by limiting the model’s parameters for her specific research. 

Finding QBS galaxies

A tidal disruption event occurs when a star gets torn apart by the gravity of a supermassive black hole. This is a transient event, meaning it produces a huge amount of light and energy. 

This energy output allows scientists to discover the size of the supermassive black holes involved. They are interested in finding galaxies where these events occur. 

TDEs are more likely to occur in Quiescent Balmer-strong galaxies, but these galaxies are hard to find. It usually takes expensive equipment and close monitoring to find them.  

Moon was able to train ML models to identify these galaxies on their own.  

“We fed the spectroscopic observations of QBS and non-QBS galaxies to the program,” she said. “The machine learning model tries to identify a relationship between each galaxy type and their photometric observations.” 

When given new examples, the models could then identify QBS from other types of galaxies. By finding QBS galaxies, scientists could find more TDEs and thus identify the sizes of supermassive black holes. 

After testing five to six other computer models, she chose two to use in her research. Moon changed the parameters of her mentor’s ML model and found her own model to adjust: 

• Random Forest – her mentor, French, used this method in her previous research. 

• K-Neighbor Classifier – Moon is the first to use this model for this kind of research.  

The ML models were able to identify 9,548 QBS galaxies. The adjusted Random Forest method had 75-96% accuracy. The K-Neighbor Classifier had 60-99% accuracy.  

Moon’s research efforts will make identifying QBS galaxies easier and cheaper. 

“Scientists have been using spectroscopy to find TDEs, but now we can do it using photometry,” she said. “This method is fast and relatively inexpensive. It also automates finding QBS galaxies which reduces the need of manpower.” 

Learn more about the physics, astronomy and material sciences department.