GB segregation substantially alters structural properties of metallic alloys, including strength, fracture mode, and corrosion resistance. A critical factor rooting the variability of GB segregation is nonetheless the atomic structure of the boundary.  GB segregation and its connection with GB structure should thus be included in modern computational strategies for polycrystalline materials design. Here, we propose an efficient and user-friendly Machine-Learning (ML) framework capable of predicting the segregation of different solute atoms at GBs  in the form of a segregation energy density, from the corresponding virgin GB atomic structure. We also show that ML provides a fresh and promising perspective to address the long-standing issue of GB structure-segregation property relationship, in the form of two  correlated atomic parameters quantifying the degree of structural and segregation symmetry respectively.