One of the production’s main stages of pipes from low-carbon and medium-carbon steel grades is heat treatment. During the hardening process, the structure of the metal changes and, as a result, the mechanical properties change. Comparing various indicators, for example, hardness, strength, plasticity, etc., it is possible to judge how successful the heat treatment regimes have been selected. Therefore, it is important to pre-establish optimal conditions in order to obtain a metal with the necessary mechanical properties. Standard approximations that allow predicting the values of mechanical properties are usually not adaptive for use in different production conditions due to the fact that in most cases they are either inaccurate or tied to a specific production unit and, as a result, are not suitable for use in other (different) conditions. The purpose of this work is to construct steel hardenability diagrams using modern machine learning methods. The choice for the study is a complex of aggregated experimental data, which includes diagrams of the decomposition of super cooled austenite, tabular values and other types of data obtained from various sources. This article describes in detail the stage of preliminary data processing, model construction and validation. Special emphasis is placed on the process of processing the initial data for modeling and comparing the fundamental features of the model with the experimental ones. The analysis of the significance of signs with real physical prerequisites is carried out in a complex. In addition, the simulation results are compared with real cal inability diagrams.