<p>Machine learning is crucial in materials design, enabling analysis of large datasets to predict properties and optimize compositions. This study used machine learning to model the relationship between the composition, solution treatment process, and comprehensive properties of high-manganese steel, aiming to optimize its properties under specified composition. By analyzing chemical composition, heat treatment, properties, and wear test conditions, a predictive model for comprehensive properties was established and used to optimize the heat treatment process, followed by experimental verification of the process before and after optimization. The results indicated that, compared to conventional single-stage treatment, the optimized step heating treatment refined austenite grains, reducing average grain size from 646.5 to 454.9 μm, and dispersed carbides more evenly at grain boundaries. Mechanical tests closely matched model predictions, with a recorded hardness of 269.5 HBW, tensile strength of 599.8 MPa, yield strength of 436.0 MPa, elongation of 12.7%, and U-notch impact toughness of 77.5&#xa0;J&#xa0;cm⁻². The optimized process significantly enhanced wear resistance, reducing wear amounts under different impact energies, with the wear mechanism shifting from microcutting to chipping wear as impact energy increased.</p> Graphical abstract <p></p>