Probabilistic material characterization is within the reach to the engineers by the virtue of machine learning methods. As the technology is rapidly evolving, its utilization calls for scrutiny and compliance to existing regulatory Certification requirements to ensure safe and efficient deployment. Physics informed machine learning is a growing approach for engineering and scientific applications. Machine Learning (ML) offers a powerful augmentation capability in building a useful physics informed complex predictive models when compared to traditional statistical methods. Limitations of this strategy need to be understood to avoid risk factors in system design choices. The term “Physics informed” simply refers to our ability to define and evaluate the contributions to the learning process by physical attributes and engineering principles. For example, conservation of mass, momentum, or energy can be accounted for in the Machine Learning training process. Several case studies in the domain of materials, such as compact-tension and compact-compression coupons were explored to characterize composite material properties. A standard approach has been established to characterize the material properties based on test data, while accounting for test anomalies (outliers) and a systematic quantification of uncertainty inherent in a physical test. This work focuses on physical test uncertainty quantification and how it impacts the accurate predictions using a ML approach. While test uncertainty has a great impact on predictions, the numerical uncertainty also needs to be accounted for ML makes it possible to process hundreds of tests coupons and issue predictions that are comprehensive, even for new material characterizations. It is expected that over time, the proper application of ML methods will produce results that are of sufficient quality and reliability that they may significantly reduce the required number of physical tests and be considered compliant with regulatory requirements.