Quality control remains the main barrier for broader adoption of Additive Manufacturing processes. Data analytics, physical process modelling, part measurement and metrological assessment, are more and more used to achieve better quality. However, there are still significant modeling, computational, and measurement challenges stemming from the broad range of the involved parameters affecting the quality of the final part. In this thesis, we focus on overcoming some of these quality-related limits. We propose a predictive modeling approach to perform porosity characterization and to determine the range of manufacturing working conditions based on a limited set of previously collected data. The proposed systematic modeling approach uses Machine Learning Gaussian Process (GP) to map the entire experimental space based on limited predetermined measured points. GP integrates a covariant function, which uses statistical bayesian inference coupled with Markov Chain to estimate model parameters, based on the collected data. These data are generated based on a proposed experimental design and CT scan image analysis protocol. Finally, and for an efficient implementation of approach, we benefit from establishing correlations between the manufacturing process conditions and the product's features, based on Key Characteristics (KCs) while considering the whole value chain in AM. These KCs are evaluated based on their importance and ordered hierarchically from a statistical point of view.