Machine learning-based classification of variable stars using phase-folded light curves

Abstract

Classifying variable stars is crucial for advancing our understanding of stellar evolution and dynamics. As large-scale surveys generate increasing volumes of light curve data, the demand for automated and reliable classification techniques continues to grow. Traditional methods often rely on manual feature extraction and selection, which can be labor-intensive and less effective for managing extensive datasets. In this study, we present a convolutional neural network (CNN)-based method for classifying variable stars using raw light curve data and their known periods. Our approach eliminates the need for manual feature extraction and preselected preprocessing steps. By applying phase-folding and interpolation to structure the light curves, the model learns variability patterns critical for accurate classification. Trained and evaluated on the All-Sky Automated Survey for Supernovae (ASAS-SN) dataset, our model achieves an average accuracy of 90% and an F1 score of 0.86 across six well-known classes of variable stars. The CNN effectively handles the diverse shapes and sampling cadences of light curves, offering a robust, automated, data-driven solution for classifying variable stars. This automated, data-driven method provides a robust solution for classifying variable stars, enabling the efficient analysis of large datasets from both current and future sky surveys.