Semi-supervised classification and clustering analysis for variable stars

The immense amount of time series data produced by astronomical surveys has called for the use machine learning algorithms to discover and classify several million celestial sources. In case variable stars, supervised approaches have become commonplace. However, this needs a considerable collection expert-labeled light curves achieve adequate performance, which is costly construct. To solve problem, we introduce two approaches. First, semi-supervised hierarchical method, requires substantially less trained than methods. Second, clustering analysis procedure that finds groups may correspond classes or sub-classes stars. Both methods are primarily supported dimensionality reduction visualization avoid curse dimensionality. We tested our with catalogs collected from OGLE, CSS, Gaia surveys. method reaches performance around 90\% all three selected stars using only $5\%$ in training. This suitable classifying main when there small training data. Our confirms most clusters found purity over respect 80\% sub-classes, suggesting type can be used large-scale variability as an initial step identify present and/or build sets, among many other possible applications.