Purpose: The prediction of the optimal bead geometry is an important aspect in robotic welding process. Therefore, the mathematical models that predict and control the bead geometry require to be developed. This paper focuses on investigation of the development of the simple and accuracy interaction model for prediction of bead geometry for lab joint in robotic Gas Metal Arc (GMA) welding process.Design/methodology/approach: The sequent experiment based on full factorial design has been conducted with two levels of five process parameters to obtain bead geometry using a GMA welding process. The analysis of variance (ANOVA) has efficiently been used for identifying the significance of main and interaction effects of process parameters. General linear model and regression analysis has been employed as a guide to achieve the linear, curvilinear and interaction models. The fitting and the prediction of bead geometry given by these models were also carried out. Graphic results display the effects of process parameter and interaction effects on bead geometry.Findings: The fitting and the prediction capabilities of interaction models are reliable than the linear and curlinear models and it was found that welding voltage, arc current, welding speed and 2-way interaction CTWD welding angle have the large significant effects on bead geometry.Research limitations/implications: The these models developed are extended to shielding gas composition, weld joint position, polarity and many other parameters which are not included in this research in order to establish a closed loop feedback control system to minimize possible errors from uncontrolled variations.Practical implications: The developed models apply real-time control for bead geometry in GMA welding process and perform the Design of Experiments (DOE) analysis steps in order to solve optimisation problems in GMA welding process.Originality/value: The interaction factors, welding voltage arc current, CTWD welding angle, also imposes a significant effect on bead geometry. With the experimental data of this study, the interaction models have a more reliable fitting and better predicting than that of linear and curvilinear models.