Variable star classification using the light curve ensemble method

Variable star classification is a crucial yet time-consuming task in astrophysics that could benefit from recent advancements in machine learning. This study explores the application of the ensemble method to enhance the accuracy and efficiency of variable star classification. We propose a multi-view classification technique called light curve ensemble (LCE), which integrates predictions from models trained on complementary light curve representations. Unlike traditional classification methods that rely solely on numeric features or single-view visual representations, our approach transforms numerical light curve data into image-based representations and applies classifications based on multiple visual perspectives. This technique leverages ensemble learning to enhance predictive performance and robustness. Experimental evaluations on a subset of the All-Sky Automated Survey for Supernovae (ASAS-SN) dataset demonstrate that the proposed LCE model outperforms conventional single-view models in variable star classification. In particular, the macro-average F1 score improves from 0.85 (best single-view baseline) to 0.89 with the proposed multi-view LCE. This gain indicates that combining complementary light-curve representations improves generalization and yields more accurate classifications across multiple variable star types. These findings highlight the potential of the proposed LCE model for advancing automated variable star classifications, paving the way for more efficient and accurate astrophysical analyses.