Abstract: Compared with wet cutting, dry cutting is classified as extreme manufacturing due to the solid-state contact at the interface, leading to stringent chip conditions that cause deterioration of the machining surface quality. To accurately predict residual stresses and roughness after the dry milling of IN718 fused cladding, a genetic algorithm-enhanced BP neural network prediction method has been proposed. An improved BP neural network prediction model has been constructed to explore the effects of milling parameters on residual stresses and roughness. The results demonstrate that the GA-BP model reduces prediction errors of residual stress and roughness from 24.8% and 18.9% to 6.6% and 10%, respectively, underscoring its advantages in fitting small sample data. Surface residual stresses and roughness have been observed to increase with higher feed rates and radial depths of cut, with spindle speed having a less pronounced effect on residual stress. The priority sequence of factors affecting residual stress has been identified as radial depth of cut > feed rate > spindle speed, while for roughness, it is feed rate > radial depth of cut > spindle speed.