<p>This study presents a comprehensive framework for optimizing the coating of thick Inconel 625 layers on low-carbon steel substrates using cold metal transfer additive manufacturing (CMT-AM). To ensure operational stability and coating integrity, this research aims to optimize critical weld bead attributes, including bead width (<i>w</i>), contact angle (<i>θ</i>), and dilution (<i>d</i>). This study proposes a hybrid method combining machine learning (ML) and multi-response optimization. This approach addresses the nonlinear relationships between process parameters (wire feed speed—<i>wfs</i> and travel speed—<i>v</i>) and bead geometry. First, neural network regression (NNR) models with Bayesian hyperparameter optimization were developed to predict bead characteristics. The NNR models achieved high predictive accuracy (R² &gt; 0.92) and low root mean square errors, outperforming linear regression (LR), second-order polynomial regression (2ndPR), support vector regression (SVR), and extra trees regression (ETR) models. Analysis of variance (ANOVA) revealed that <i>wfs</i> is the dominant factor, contributing over 70% to the variation in bead width (<i>w</i>) and dilution (<i>d</i>). Subsequently, a multi-objective neural network algorithm (MONNA) was employed to generate a Pareto front of non-dominated solutions, identifying trade-offs between geometric formability and metallurgical quality. Grey relational analysis (GRA) was then applied to select the best compromise solution from the Pareto set. The optimal process parameters were identified as <i>wfs</i> = 9&#xa0;m/min and <i>v</i> = 53&#xa0;cm/min. Experimental validation confirmed that these conditions produce high-quality thick coatings with uniform thickness, sound metallurgical bonding, and defect-free microstructures. This work demonstrates the efficacy of data-driven modeling combined with multi-criteria decision-making for process optimization in wire arc additive manufacturing.</p>