Most materials are heterogeneous, depending on the scale at which they are observed. Therefore, it can be said that the mechanical properties of the material are largely dependent on the heterogeneous microstructure, i.e., on the mutual interaction of different constituents at the microstructural level. In recent times, there has been an accelerated development of advanced multilevel numerical algorithms with the help of which is possible to define a direct connection between the microstructure and the constitutive behavior of the material at the macrolevel. In this way, it is possible to develop a material with optimal properties by manipulating the distribution, volume fraction and properties of microconstituents. In this paper, nodular cast EN-GJS-400-18-LT produced by Tundish and Inmould casting techniques was considered. An analysis of the existing microstructures was carried out with the help of the described distributions according to the diameter of the graphite nodules and their mutual distance. The most accurate distribution for both variables was determined and the microstructure of ductile iron was generated for both methods of casting. The values of the material parameters of the ferrite matrix of nodular cast iron were calculated using the reduced homogenization method. With the help of the k-means clustering algorithm based on machine learning, the material points of the microstructure are grouped into elements with similar properties, in which the stress and strain values are equal. All of the above allows the method to discretize the computational model with a relatively small number of clusters, which enables a significant saving of computation time. After clustering, the Lippmann-Schwinger equation is solved incrementally. On the basis of numerical simulations, the coefficients of complete polynomials of the second and third order that define the elastoplastic properties of the ferrite matrix according to Swift's law of hardening were determined, by means of which the actual values of the material parameters of the ferrite matrix were determined. The accuracy of the method itself was demonstrated by comparing parameters calculated by numerical simulations with real values taken from the literature.