Jennifer Nakano was born and raised on the Big Island of Hawaii and graduated from Waiakea High School. She is attending the University of Hawaii at Manoa, pursuing computer science with a minor in mathematics. Jennifer is interested in data science, machine learning, and AI because of its adaptability to various fields and capability to make a difference in the world. Her goals are to work in the field of data science and work somewhere in Hawaii so that she can stay close to her family. Jennifer enjoys baking, playing video games, and spending time with her friends and family.
Home Island: Big Island of Hawaii
High School: Waiakea High School
Institution when accepted: University of Hawaii at Manoa
Classifying Variable Stars in the Large Magellanic Cloud through Machine Learning
Project Site: Gemini Observatory, Hilo, HI
Mentors: Clara Martinez-Vazquez & Ricardo Salinas
Project Abstract:
As the number of astronomical surveys increase with technological advancements, the data collected from these surveys grows exponentially. Recently, a research team at Gemini carried out a stellar variability survey in the Large Magellanic Clouds using the Dark Energy Camera associated with Gemini’s parent organization, NOIRLab. This survey has collected time-series data of about 5.5 million objects, and scientists now face the daunting task of determining which of these objects are variable stars and classifying what type of variable stars (delta Scuti, RR Lyrae, Cepheids, eclipsing binaries) they are. These classifications are based on observed stellar light curves and derived periods, along with their positions in color-magnitude diagrams generated from the survey. In hopes of improving the efficiency in characterizing the variable stars in this survey, we use machine learning models to automate the task of processing all of the incoming data. The machine-learning models XGBoost and the Histogram-based Gradient Boosting Classification Tree look for patterns such as variability in stellar magnitudes, and are trained to optimize the detection of variable stars without introducing contamination from false positives (non-variable stars that are incorrectly classified as variable stars). K-Nearest Neighbors and template fitting utilizes the period, luminosity, magnitude relation as well as folded light curves to distinguish the different variable star types. After the two phases of classification, the machine learning models achieve < 80% accuracy.