Dissimilar metal welding, particularly between mild steel and AISI 304 stainless steel, has gained significance in various industrial applications due to its unique combination of properties. However, dissimilar welding often experiences a higher failure rate due to the heterogeneity of the weld at the microscale. Differences between mild steel and AISI 304 stainless steel can lead to challenges such as residual stresses, brittle intermetallic compounds, and poor weld quality. Since welding process parameters play a crucial role in achieving the desired weld quality, researchers commonly employ optimization techniques to enhance mechanical properties and improve weld performance. In this study, an ANN using a Backpropagation Neural Network (BPNN) model was built to predict the tensile strength and FZ hardness in GTAW of dissimilar welding. The BPNN model was further utilized to optimize welding parameters, enhancing overall weld quality. The experiments were designed using Taguchi’s L9 orthogonal array, with welding current and welding filler as the selected process parameters. The Taguchi method has shown that AWS 70S-6 welding filler and 140 A was the optimal combination, while the optimal setting presented by the BPNN model prediction was AWS308L and 110 A. The Schaeffler diagram has predicted a martensitic FZ phase for AWS 70S-6 welding, while it was a duplex phase for AWS 308L and AWS 309L welding. The BPNN model of an 80/10/10 split with a hidden layer of 5 neurons has shown higher accurate prediction capability than Taguchi’s optimal prediction, resulting in lower prediction error.