Abstract: Due to the complexity of interactions among multiple process parameters, accurately predicting the geometric shape of formed weld beads in wire arc additive manufacturing (WAAM) remains a challenge. Traditional single-parameter models or empirical formulas often fail to fully capture this nonlinear, highly coupled process-morphology relationship, which limits the effectiveness of process optimization and quality control. To address this issue, a universal weld bead geometry prediction model based on curvature features was proposed, which could adapt to different combinations of process parameters and achieve an accurate description of the cross-sectional shape of deposited layers. Through systematic experiments, a comprehensive dataset including the weld curvature, the weld contact angle, and key process parameters was obtained. On this basis, multiple machine learning prediction models, including the XGBoost model and the support vector machine (SVM) model, were developed. By providing a universally applicable strategy for predicting the deposition geometry and further optimizing the process parameters, theoretical and data references are provided for the subsequent process parameter optimization and intelligent control strategy formulation of WAAM.