Nickel-based superalloys are unique materials with complex doping that demonstrate excellent resistance to mechanical and chemical degradation. Over a long period of use in industry, a variety of information has been accumulated about their possible chemical compounds and features corresponding to a certain composition. One of the main service properties of the alloy is the tensile strength (σ). For a more convenient comparison of the characteristics of the alloys with different chemical compositions, the temperature and holding time of the metal during testing are often combined into a complex Larson–Miller parameter (PLM). The availability of experimental or simulated data on the alloys' properties in the entire range of temperatures and exposures would significantly expand the possibilities of the alloy applications and would allow a more accurate evaluation and comparison of the alloys. In this work, we use a machine learning method for modeling the properties of the alloys according to their composition. A Bayesian regularized artificial neural network was engaged to simulate missing tensile strength values for 278 superalloys. Special data preprocessing and the use of an ensemble of networks during training reduced the model error. Comparison of the predicted and experimental data showed excellent convergence. The method made it possible to obtain enough data to approximate the relation (Figure presented.) with a sigmoid function. The slope coefficient is considered as a quantitative expression of the thermal stability of the superalloys.