Electrochemical　CO2　reduction　to　multi-carbon　products　is　pivotal　for　sustainable　energy　and　carbon　neutrality,　yet　bimetallic　catalyst　design　remains　hindered　by　unclear　structure–activity　relationships.　We　present　a　high-throughput　framework　integrating　density　functional　theory　(DFT)　and　machine　learning　(ML)　with　a　“classification-regression”　dual　model.　Screening　435　bimetallic　catalysts　on　nitrogen-doped　graphene　(M1M2@Gr),　we　identified　37　candidates.　The　XGBoost　classifier　achieved　0.911　accuracy　(AUC　=　0.895),　with　SHAP　analysis　highlighting　M1-C　bond　length　as　the　critical　descriptor.　Regression　models　precisely　predicted　stability　and　intermediate　energies.　DFT　validation　revealed　Fe_Co,　Fe_Ir,　and　Rh_Re@Gr　as　top　performers,　exhibiting　ultralow　rate-determining　barriers　(＜0.5　eV)　due　to　dynamic　electron　“acceptance-donation”　synergy　between　metal　sites　and　CO.　This　work　establishes　an　ML-DFT　paradigm　for　rational　catalyst　design.　©　2025　Elsevier　Ltd