• Machine learning-driven deep potential molecular dynamics (DPMD) accurately predicts the local structure and thermophysical properties of Mg–La liquid alloys, overcoming the accuracy-efficiency dilemma of traditional simulation methods.
• The effect of increasing Mg content on the first coordination shell mirrors the effect of increasing temperature, providing a unified understanding of structural evolution in Mg–La melts.
• A medium-range ordered structure, indicated by a pre-peak in the structure factor, is prominent in the 80at% Mg alloy and vanishes at elevated temperatures, revealing temperature-sensitive ordering.
• The study establishes a reliable database of density, self-diffusion coefficients, and shear viscosity for Mg–La alloys, and evaluates mixing enthalpy and elemental activity at 1200 K, offering new guidance for alloy design and processing.