Abstract <p>The article considers the development of a setup for studying the thermo-EMF of metallic materials from industrially produced alloys for various purposes. The setup allows measuring the thermo-EMF using thermocouples made of the alloys under study, synchronously with temperature measurements. An integral method with subsequent mathematical processing to calculate the Seebeck coefficient is used for the measurement. The error of the mathematical processing is estimated, and the values of the Seebeck coefficient are verified against the data of other studies in which an accurate differential measurement method was used. Temperature curves of the Seebeck coefficient are obtained for 73 grades of industrial alloys, including alloyed steels, nickel-based alloys, titanium, and some refractory metals. Analysis of the obtained data using machine learning methods makes it possible to identify generalized features of thermo-EMF changes for the studied groups of materials and classify them by the type of thermo-EMF curves. A model for predicting the Seebeck coefficient for new materials based on the experimental data regression is developed and trained. To increase prediction accuracy, the measured values are used not only as a training model response, but also as additional features along with data on the alloy chemical composition.</p>