ART (Adaptive Resonance Theory) networks were invented in the 1990s as a new approach to the problem of image classification and recognition. ART networks belong to the group of resonance networks, which are trained without supervision. The paper presents the basic principles for creating and training ART networks, including the possibility of using this type of network for solving problems of predicting and processing measurement data, especially data obtained from geodesic monitoring. In the first stage of the process of creating a prediction model, a preliminary analysis of measurement data was carried out. It was aimed at detecting outliers because of their strong impact on the quality of the final model. Next, an ART network was used to predict the values of the vertical displacements of points of measurement and control networks stabilized on the inner and outer walls of an engineering object.