Hierarchical Classification of Variable Stars Using Deep Convolutional Neural Networks

TL;DR

This paper demonstrates that hierarchical deep neural networks, specifically CNNs and RNNs, can effectively classify variable stars with high accuracy using light curves and periods, improving classification of smaller classes.

Contribution

It introduces a hierarchical classification approach with deep neural networks for variable star classification, enhancing accuracy especially for rare subclasses.

Findings

Achieved 98% accuracy for class classification.

Achieved 93% accuracy for subclass classification.

Hierarchical approach improves classification of small classes.

Abstract

The importance of using fast and automatic methods to classify variable stars for large amounts of data is undeniable. There have been many attempts to classify variable stars by traditional algorithms like Random Forest. In recent years, neural networks as classifiers have come to notice because of their lower computational cost compared to traditional algorithms. This paper uses the Hierarchical Classification technique, which contains two main steps of predicting class and then subclass of stars. All the models in both steps have same network structure and we test both Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). Our pre-processing method uses light curves and period of stars as input data. We consider most of the classes and subclasses of variable stars in OGLE-IV database and show that using Hierarchical Classification technique and designing appropriate preprocessing can increase accuracy of predicting smaller classes, ACep and T2Cep. We obtain an accuracy of 98% for class classification and 93% for subclasses classification.