Enhancing Galaxy Morphology Classification using Ensemble Learning

In recent decades, due to progress made in the field of astronomy and the development of specialized equipment, large-scale sky surveys have generated immense amounts of data. To address this issue, astronomers have utilized the power of crowdsourcing provided by the Galaxy Zoo project for the morphological classification of galaxies. However, with the upcoming generation of surveys, relying on crowdsourcing is not optimal and the adoption of automated strategies for the morphological classification of galaxies is essential. The aim in this work is to design a model focused on the classification of galaxies, exploring average weighted ensemble learning and comparing the outcomes with baseline models using a dataset from the astronomical DESI Legacy Imaging Surveys (DESI), which consists of approximately 18,000 images from 10 broad galaxy types. The results reveal that employing ensemble learning outperforms the baseline models and produces better results compared to them. In this work, models such as ResNet-50, Ef-ficientNetV2S and DenseNet-121 are employed to create the ensemble, with the ensemble achieving a test-set accuracy of 89%, compared to ResNet-50, DenseNet-121 and EfficientNetV2S achieving 86%, 87% and 88%, respectively.