Detecting Indicators of Horizontal Collusion in Public Procurement with Machine Learning Methods

Abstract

Improvement of procurement procedures and their digitization help prevent and identify cartels, but at the same time lead to the emergence of new anticompetitive schemes. In this paper we focus on electronic auctions, which have become the main method of public procurement in Russia in recent years. As e-auctions provide access to many big government orders; the incentives for bidders to join anti-competitive agreements are increased. Therefore, the development of methods to detect bid rigging at electronic auctions is of high practical importance. The aim of this work was to develop a method for detecting signs of horizontal collusion at an auction. We use machine learning methods to train classifiers that predict the presence or absence of cartel in electronic auctions, depending on the distribution of bidders, the time of submission of applications, the duration of the auction and the number of participants. Variables for the model were selected on the basis of distribution plots built for sample of cartels and random sample. The study is based on data from public procurement Web portal and the information about bid rigging from cases of the Federal Antimonopoly Service. The results showed that the Random forest model most accurately predicts the detection of the cartels on electronic auctions. The accuracy of the prediction is 84%, and the recall and precision of the model are 83 and 87%, respectively. The most significant variables for the classification are the level of price reduction, the difference in the time of application filing of participants and the value of the maximum starting price of contract.