Unsupervised Machine Learning for Classifying Variability in X-ray Light Curves: Case Study on GRS 1915+105

Abstract

X-ray astronomy provides a unique window into high energy astrophysical processes, with light curves serving as key tools for probing the temporal variability around compact objects such as black holes and neutron stars. Traditionally, analyzing such variability requires manual inspection and expert defined features, which become increasingly infeasible with the growing volume of data from modern X-ray missions. In this work, we explore the potential of unsupervised Machine learning for classifying X-ray light curve morphologies. We demonstrate our approach using simulated light curves inspired by the variability classes of the micro quasar GRS 1915+105, a well-studied black hole X-ray binary known for its rich and diverse temporal behavior. By selecting representative patterns from the GRS 1915+105 variability zoo, we train and evaluate clustering algorithms on their ability to group light curves. Our results show that unsupervised learning methods can successfully distinguish between different morphological classes without prior labels, highlighting the promise of Machine Learning in automating and scaling, the time domain analysis in X-ray astronomy.