Qingyang Ye, Yiran Cheng, Liang Cao
High-entropy alloy (HEA) nanocatalysts hold great promise as heterogeneous catalysts, yet their rational design remains a formidable challenge due to complex atomic arrangements and a vast compositional space. Here, we present a density functional theory (DFT)-trained machine learning cluster expansion (ML-CE) framework for high-throughput screening of quinary HEA compositions sampled systematically at 5% intervals. Coupled with Metropolis Monte Carlo (MMC) simulations, the framework resolves temperature- and atmosphere-dependent surface segregation and then predicts equilibrated structures and site-averaged turnover frequencies. For Ir–Pd–Pt–Rh–Ru octahedral nanoparticles in the oxygen reduction reaction (ORR), the screening predicts Pt- and Rh-rich, Pd- and Ru-lean formulations that could deliver up to ∼12-fold higher ORR activity than commercial Pt/C. Statistical analysis showed that highly active sites are dominated by first-nearest-neighbor triplets composed exclusively of Pt and Pd atoms. The framework enables rapid composition–activity mapping across multicomponent alloys, links surface segregation to catalytic performance, and derives transferable design principles accounting for segregation dynamics, adsorbate-binding energetics, and local coordination, offering a practical strategy for rational HEA catalyst discovery.