Wire-arc additive manufacturing (WAAM) is an advanced technique for fabricating large metal components through layer-by-layer material deposition using arc welding methods. This study focused on optimizing the WAAM process by employing machine learning models to predict and control bead geometries, specifically bead height (BH) and bead width (BW), while ensuring consistent height increments in multibead walls. Based on CMT technology in cold metal transfer experiments, linear regression models achieved high accuracy in predicting BH and BW. Analysis of variance results highlighted the considerable influence of voltage (V) and travel speed (TS) on bead geometries. For multibead wall characteristics, polynomial regression models incorporating non-linear terms, such as travel speed (TS²) and dwell time (Dt²), were developed to predict height (H) and waviness (W). Various optimization metrics were employed to balance the trade-offs between H and W for identifying optimal welding conditions that achieved the target H while minimizing W. A notable innovation of this research is the optimization of dwell time (Dt) for each layer to achieve a linear incremental H profile, minimizing W and ensuring consistent layer quality.