Anastassiya Ponomaryova

Astronomical Classification

Started June 4, 2025

Abstract or project description

This project focuses on the classification of astronomical objects using deep learning techniques applied to photometric time-series data. Astronomers collect huge amounts of data from telescopes, and therefore, manual astronomical classification is time-consuming, difficult, and prone to errors. Automated classification is essential for identifying stars, galaxies, supernovae, and other transient objects efficiently. The dataset includes light curves with features such as Modified Julian Date (MJD), flux, flux error, passband, and detection flags, supplemented by metadata indicating object type. Several models were developed and evaluated, including LSTM networks, dense neural networks, and convolutional neural networks (CNNs). Data preprocessing involved scaling, categorical encoding, and sequence padding to align variable-length light curves. Experiments were conducted on datasets spanning multiple passbands and object classes.