Classification of Astronomical Objects in the Galaxy M81 using Machine Learning Techniques II. An Application of Clustering in Data Pre-processing

摘要

Identifying objects with a certain class in the current data in astronomy are challenging. In this study, we explored the methods to identify globular cluster candidates from a pool of astronomical objects in the galaxy M81. First, we developed a method to automatically cross-match the data. This process was done by manually overlayed the imaging data in the previous study. The process also eliminated the data points that only appear in only one or two filters, which indicates that they are artifacts. Next, we used the Expectation Maximization (EM) clustering technique to label the training dataset with classes and to reduce the use of humans in the preprocessing process. Our results show that the data can be clustered into 12 clusters, which can be grouped into 6 groups of astronomical objects with similar morphological structures. When using these 6 groups of data to build classification models, we found that the prediction accuracies have improved significantly. In the case of Random Forest, the accuracy has improved from 79.9% to 90.57% and from 67.1% to 91.59% for Multilayer Perceptron. Moreover, when using the model built from those data to analyze the unseen dataset, the results also show that the model can categorize the objects into classes with characteristics close to those in astronomy. However, this model still cannot fully separate globular clusters from foreground stars and background galaxies due to the similarities in their photometric properties.