As one of the most popular additive manufacturing techniques for inorganic materials, powder bed fusion (PBF) of alloy materials has shown industries flexibility and rapidness in manufacturing novel and complex geometries. Modeling and simulation of the PBF aim to complement the current time and cost expensive trial-and-error principle with an efficient computational design tool. Nevertheless, due to the sophisticated and interactive nature of underlying physics, deviations generated by incomplete physical interactions and uncertain parameterization can make the simulation results less robust. Concluding simulation-based phenomenological laws/relations also suffer from insufficient size of dataset, as it generally requires a vast computational resources and time [1]. In this regard, there is great interest in using machine learning (ML) models to perform the data-driven analysis. In this work, we perform the multiphysics PBF simulations considering multiple factors that can influence the properties of the manufactured parts, such as the thermal conductivity and capacity of the powder bed and their temperature dependency. Uncertainty of physical and processing input parameters are considered via random sampling within the physical limits. Data-driven analyses based on the batched parallel simulations are performed to reveal the sensitivity of the input parameters to the output properties of interest. The properties include the melt pool geometry, porosity, and residual stress. The inference of the input parameters under the experimental conditions are then conducted. We further verify the dimensionless scaling law of melt pool control [2] via law of large numbers (LLN), which presents the convergence to the theoretical limit derived from thermal analysis.