📝 SummarXiv

Abstract

Star formation (SF) studies are benefiting from the huge amount of data made available by recent large-area Galactic plane surveys conducted between 2 {\mu}m and 3 mm. Fully characterizing SF demands integrating far-infrared/sub-millimetre (FIR/sub-mm) data, tracing the earliest phases, with near-/mid-infrared (NIR/MIR) observations, revealing later stages characterized by YSOs just before main sequence star appearance. However, the resulting dataset is often a complex mix of heterogeneous and intricate features, limiting the effectiveness of traditional analysis in uncovering hidden patterns and relationships. In this framework, machine learning emerges as a powerful tool to handle the complexity of feature-rich datasets and investigate potential physical connections between the cold dust component traced by FIR/sub-mm emission and the presence of YSOs. We present a study on the evolutionary path of star forming clumps in the Hi-GAL survey through a multi-step approach, with the final aims of (a) obtaining a robust and accurate set of features able to well classify the star forming clumps in Hi-GAL based on their evolutionary properties, (b) establishing whether a connection exists between the cold material reservoir in clumps, traced by FIR/sub-mm emission, and the already formed YSOs, precursors of stars. For these purposes, our designed experiments aim at testing whether the FIR/sub-mm properties related to clumps are sufficient to predict the clump evolutionary stage, without considering the direct information about the embedded YSOs at NIR/MIR. Our machine learning-based method involves a four-step approach, based on feature engineering, data handling, feature selection and classification. Our findings suggest that FIR/sub-mm and NIR/MIR emissions trace different evolutionary phases of star forming clumps, highlighting the complex and asynchronous nature of the SF process.