Using Machine Learning to Find New Members of the Pleiades

Abstract

An open star cluster is a group of gravitationally bound stars that move together through space and have the same origin. One of the most famous star clusters, which can be seen with the unaided eye, is Pleiades. In order to accurately model clusters’ formation and evolution, scientists need to know exact cluster members to include them into analysis. Unlike the stars that are close to the cluster core, it is hard to relate the stars that are farther from the cluster center and can therefore be confused for field stars. In this thesis, we find Pleiades member candidates among stars with unknown membership using Machine Learning. Here we show that spectral data alone is not enough for clear membership determination, although combined with stars’ positions and velocities, it produces valid results. Our 22 suggested Pleiades member candidates have positions, velocities, abundances and atmospheric parameters similar to the Pleiades stars. Features with more predictive power are positions and abundances Fe=H, M=H and C=Fe. The model relying on spectral features has been able to find a lot more stars with chemical composition similar to the Pleiades. The fact that some of these predicted stars are too far away (more than 20 pc from the cluster center) proves that spectral information alone is not discriminative enough to isolate the members of one particular cluster. Still, it is very useful to separate Pleiades member candidates from field stars, since the precision of the model on the test dataset is 0.957. Features that are more important for the prediction are N=Fe, C=Fe and Teff . The results obtained in this thesis will be very useful for large future sky surveys. Having many stars as possible cluster members, our model will help to carefully reduce their number for a detailed membership study.