Machine Learning with Astronomical Data: 
Variable Object Classification

Author: Sohum Berry

Affiliation: Menlo School, California

​

Abstract:​​​​

 

In the current era of telescope development, new techniques must be used to make sense of the vast amounts of data that is being collected. Machine learning models present a neat solution for classifying variable objects by their measured features. The aim of this paper is to determine which machine learning model can more accurately determine the class of variable objects from the Gaia Data Release 3. The models tested are Logistic Regression Classification, Naïve Bayes Classification, Random Forest Classification, and finally Neural Networks. Random Forest Classification had the most accuracy and adjustability for this use case, however Neural Networks are also a viable option. When using a Random Forest Classification to reclassify Gaia low confidence predictions, the models largely agree on classifications of long period variable and solar like objects, however the results are not too consistent for other classifications.

​

Keywords: astronomy, variable objects, classification, machine learning, random forest​

​​

ISSN 3069-8200