The defects introduced by additive manufacturing (AM) are deemed as an essential factor to determine the service performance of near-net-shape metallic components since these defects can act as the stress concentrator and resultant crack initiator. Together with traditional post-mortem fatigue experiments on AMed specimens, high-resolution synchrotron radiation and microfocus X-ray computed tomography was also collaboratively employed to acquire rich defect data in three dimensions and to feed the subsequent probabilistic statistical model. In this respect, the support vector machine (SVM), a well-defined machine learning (ML) model, was particularly selected to perform the regression analysis on defects and experimental lifetime for the service performance assessment. The location, morphology, dimension, population and coupled effect on the fatigue degradation level were quantitatively characterized by using machine learning. This work can provide a significant reference for integrity structural assessment of AMed metal parts.