The process of constructing machine learning models for predicting the microstructure of pipe steels after continuous cooling involves the assembly and preparation of data, the source of which are thermokinetic decay diagrams of supercooled austenite. Statistics of intermediate and final data, as well as algorithms for their transformation are given. Estimates of machine learning models for selected microstructures are considered. A method for generating data in conditions of a small sample and the introduction of an estimated feature of grain size are proposed. Validation of the models and interpretation of the significance of the features were carried out. The practical use of models for constructing thermokinetic diagrams of austenite decay and analysis of simulation results is shown.