Robotic welding and additive manufacturing (AM) processes have an intricate design space influenced by numerous configurable process parameters. Currently, the precise impact of each parameter or a combination of them on the variability and dimensions of deposited material is unclear due to the stochastic nature of the process, which is affected by factors like arc stability, temperature gradients and other in-process changes. In AM and various cases of welding like cladding, quantifying these variations is necessary for developing path planning strategies that produce components without defects. This study presents a framework that automates process data collection and scanning of the weld bead and analysis of the point cloud, based on the design of experiments principals towards building representative machine learning models. In comparison to alternative approaches, this framework incorporates spatial variation along the deposited length by utilising location-based binning of measurements, thereby enabling more detailed analysis of various deposition stages including arc ignition and extinction regions. The framework is tested with single pass bead-on-plate weld beads deposited with different process parameters followed by spatial–temporal matching. Variations were noted in relation to travel speed and welding current when subjected to identical heat input values. Machine learning models for prediction of height and width account for non-linearities and are validated with additional experimental data. These models have demonstrated a high degree of accuracy in predicting in-process variations within the deposited material.