Machine Learning to Classify Young Star’s Time Series

We present a variability and morphological classification study of TESS light curves for T Tauri star candidates in the Orion, IC 348, γ Velorum, Upper Scorpius, Corona Australis, and Perseus OB2 regions. We propose eleven morphological classes linking brightness variation behaviors with possible physical or geometric phenomena present in T Tauri stars, and develop a supervised machine learning algorithm to automate the classification among these. Our algorithm optimizes and compares the true positive rate (recall) among k-nearest neighbors, classification trees, random forests, and support vector machines. This is done characterizing light curves with features depending on time, periodicity, and magnitude distribution. Binary and multiclass classifiers are trained and interpreted in a way that allows our final algorithm to have single or mixed classes. In the testing sample, the algorithm assigns mixed classes to 27% of the stars, reaching up to five simultaneous classes. A catalog of 3672 T Tauri star candidates is presented, along with their possible period estimations, predicted morphological classes, and visually revised ones. The cross-validation estimated performance of the final classifiers is reported. Binary classifiers surpass multiclass recall values for classes with less representation in the training sample. Support vector machines and random forests classifiers obtain better recalls. For comparison, another performance estimation of the final classifiers is calculated using the revised classes of our testing sample, indicating that this performance excels in singled classed stars, which happens in about 75% of the testing sample.