The purpose of the article is to substantiate the application of the authors’ approach and methodology based on a combination of machine learning technologies and the construction of directed graphs with their subsequent clustering for a systematic study of the quantitative and qualitative characteristics of the public procurement market and the behavior of agents in this market. As a result of the study, regional and sectoral factors influencing the relationship between agents of the public procurement market were identified. Such factors were not previously identified, and were determined only thanks to the combination of machine learning technologies and the theory of networks and graphs proposed by the authors. Another result of the study-the models of relationships in public procurement market are systematized in the authors’ interpretation, integrating the macroeconomic situation in the market and the marketing strategies of market players. Such stable patterns of behavior of agents of the public procurement market as "isolation", "conservatism" and "mobility" were identified, and it was substantiated that the isolated or conservative behavior of market players increases the likelihood of corrupt conspiracies. All of the above was not systematically studied before. So, it has scientific novelty and high practical significance. The research contributed to the increment of scientific knowledge in application of the theory of networks and graphs, in problems of state regulation of the economy, counteraction to the monopolization of markets and encouraging competition. The practical results of the work are related to the generation of recommendations to the Russian authorities, regulators of the public procurement market and bidders on the choice of effective market behavior strategies.