Multiple Machine Learning as a Powerful Tool for Star Cluster Analysis

Abstract

This work proposes a multiple machine learning method (MMLM) aiming to improve the accuracy and robustness of the analysis of star clusters. The MMLM performance is evaluated by applying it to the reanalysis of an old binary cluster candidate—comprised of NGC 1605a and NGC 1605b—found by D. Camargo (2021; hereafter C21). The binary cluster candidate is analyzed by employing a set of well-established machine learning algorithms applied to the Gaia-EDR3 data. Membership probabilities and open clusters (OCs) parameters are determined by using the clustering algorithms pyUPMASK, ASteCA, k-means, GMM, and HDBSCAN. In addition, a KNN smoothing algorithm is implemented to enhance the visualization of features like overdensities in the 5D space and intrinsic stellar sequences on the color–magnitude diagrams. The method validates the clusters’ previously derived parameters; however, it suggests that their probable member stars are distributed over a wider overlapping area. Finally, a combination of the elbow method, t-SNE, k-means, and GMM algorithms groups the normalized data into six clusters, following C21. In short, these results confirm NGC 1605a and NGC 1605b as genuine OCs and reinforce the previous suggestion that they form an old binary cluster in an advanced stage of merging after a tidal capture during a close encounter. Thus, MMLM has proven to be a powerful tool that helps to obtain more accurate and reliable cluster parameters, and its application in future studies may contribute to a better characterization of the Galaxy’s star cluster system.

Keywords

• Algorithms
• Open star clusters
• Clustering
• Galaxy structure
• Star clusters