Typical materials development route goes through exploring the cycle of materials processing, structure and the properties. The route that is traditionally used is through experimental characterization and is costly. In this paper, a new approach that involves the use of machine learning to identify and reduce the number of governing factors at each length scale and process, is presented. Aluminum alloy AlSi10Mg is used to study the influence of microstructural heterogeneity and defects-voids distribution on the evolution of plastic anisotropy and failure under uniaxial tensile deformation. For this purpose, ASTM-E8M subsize tensile specimens are printed using five different build directions by powder-bed fusion SLM process. The as-built specimens are subsequently scanned using 3D X-ray Nano-CT scanner to characterize the average volume fraction, average size, and shape-size distribution of pores. The microstructure and crystallographic texture of the as-built specimens are characterized by performing electron backscatter diffraction (EBSD) measurements. Quasi-static, room temperature tensile tests are performed and a digital image correlation (DIC) system is used to measure the surface strain evolution during tensile deformation. The proposed machine learning based numerical framework is employed to study the tensile fracture behavior to understand the relationship between defects-voids shape and distribution, the observed mechanical behavior, the evolution of plastic anisotropy and local failure strain.