T-Tauri stars: Identification by explainable machine learning and investigation into the effects of starspots

Abstract

A naive Bayes classifier for identifying Class II YSOs has been constructed and applied to a region of the Northern Galactic Plane containing 8 million sources with good quality Gaia EDR3 parallaxes. The classifier uses the five features: Gaia G-band variability, WISE mid-infrared excess, UKIDSS and 2MASS near-infrared excess, IGAPS Hα excess and overluminosity with respect to the main sequence. A list of candidate Class II YSOs is obtained by choosing a posterior threshold appropriate to the task at hand, balancing the competing demands of completeness and purity. At a threshold posterior greater than 0.5 my classifier identifies 6504 candidate Class II YSOs. At this threshold I find a false positive rate around 0.02 per cent and a true positive rate of approximately 87 per cent for identifying Class II YSOs. The ROC curve rises rapidly to almost one with an area under the curve around 0.998 or better, indicating the classifier is efficient at identifying candidate Class II YSOs. When comparing the results to published catalogues from other young star classifiers, between one quarter and three quarters of high probability candidates are unique to each classifier, indicating no single classifier is finding all young stars. The Colour-Magnitude Diagrams of young star clusters exhibit a scatter in their members that may naively be interpreted as a scatter in age. It may instead be due to different physical characteristics of the stars. One possible cause is a range of magnetic field strength. This would inhibit convection and give rise to cool starspots, leading to a range of radius inflation as the young stars contract more slowly towards the main sequence. A spectroscopic technique is trialled for the detection of two temperatures in the spectra of Weak-lined T-Tauri stars, a potential indicator of a spotted surface.