Globular cluster classification in galaxy M81 using machine learning techniques

Abstract

Copyright © JCSSE 2020 - 17th International Joint Conf. on Computer Science and Software Engineering. Globular clusters are very important in astronomy since they can be used to study the process of galaxy formation and evolution. With the exponential growth of data in astronomy, it is currently inefficient to classify globular clusters from the other types of astronomical objects by humans. In this study, we explored the possibility of using machine learning in globular cluster classification to replace the classification by humans. We selected five standard classification methods including k-NN, Random Forest, SVM, Neural Network, and Decision Tree. All models were built and tested by using the Weka software with datasets from a galaxy M81. Our experiments showed that k-NN, Random Forest, and SVM are the best approaches for globular cluster classification, with 97.7% accuracy, 97.8% precision, 97.7% recall, and 97.7% F-measure. Finally, when we applied these models to an unseen dataset to predict new globular cluster candidates, we acquired 6.32% success rate compared to 30% success rate by humans. This suggests that machine learning techniques can be applied to globular clusters classification. However, our models need to be improved to achieve a higher success rate to replace the classification by humans.