Towards Automated Galaxy Morphology and Redshift Estimation with a Dual-Pipeline Machine Learning Framework Using GalaxiesML

Unraveling the mysteries of our universe requires precise understanding of galactic distances and redshift estimation plays a pivotal role in this pursuit. Our proposed study explores the use of advanced machine learning for redshift prediction, utilizing the GalaxiesML dataset, which contains photometric and morphological information for 286,401 galaxies from the Hyper Suprime-Cam Survey. We propose a two-stage approach: first, classifying galaxies into low, medium, and high redshift categories using an XGBoost classifier, achieving 89% accuracy; and second, predicting continuous redshift values with a Deep Neural Network (DNN), reaching an $R^{2}$ score of 0.8805. To improve model interpretability, SHAP (SHapley Additive Explanations) analysis was applied, revealing that features such as the Sérsic index, half-light radius, and gr color index significantly influence redshift predictions. Further astrophysical analysis, including star formation rate, luminosity, absolute magnitude, ellipticity, and surface brightness, was conducted to understand the physical factors driving redshift variation. This integrated framework not only enhances redshift estimation accuracy but also deepens insight into galaxy structure and evolution, demonstrating the powerful synergy between machine learning and astrophysical research.