Public procurement plays a crucial role in economic development, accounting for a significant portion of government expenditures worldwide. However, the sector remains highly vulnerable to corruption, inefficiencies, and financial mismanagement, leading to severe economic and social consequences. Corruption in public procurement not only undermines fair competition but also results in inflated costs, poor-quality services, and weakened public trust in governmental institutions. The digitization of legal algorithms presents a promising solution to these challenges. By leveraging artificial intelligence (AI), big data, blockchain, and automated decision-making systems, governments can strengthen compliance, detect fraudulent activities, and enhance oversight in public procurement. The system for detecting fraudulent activities in public procurements and tenders is proposed. The system gets data from the open database of companies, their activities, tenders, court cases, as well as tax and legal information. Using an algorithm, the system can detect any activities with corruption threats. After analyzing hundreds of tenders, a dataset was formed. This dataset of tenders, companies and their activity will be used for training a machine-learningbased algorithm to further enhance the analytical capabilities of the system.
Corruption in public procurement is one of the most serious problems of the modern public administration system. It undermines public trust, limits economic development and leads to inefficient spending of budget funds. At the same time, existing control tools do not allow for timely and comprehensive
monitoring of possible corruption schemes and connections between procurement participants due to the large volume of data and the complexity of analyzing the relationships. As a result, violations are often identified after the completion of procedures, which significantly reduces the possibility of restoring justice and efficient use of resources.
One of the industries that is particularly susceptible to corruption risks is the construction sector.
This is due to the large volumes of funding and the complexity of projects. According to studies by
the European Commission and the Organization for Economic Cooperation and Development
(OECD), up to a fifth of the European Union's GDP is spent through public procurement, and losses
from corrupt practices reach 20-25% of the allocated funds. The most common forms of corruption in
procurement include bribery, collusion between participants, conflicts of interest and hidden
lobbying. Recent Eurobarometer surveys confirm the prevalence of the problem: a significant share
of entrepreneurs consider corruption a serious barrier to participation in tenders [
To address these issues, European countries are actively implementing digital technologies and legislative measures aimed at increasing transparency. Examples of successful initiatives are laws on access to information - the Freedom of Information Act in the UK. The law allows the public to access government data, significantly reducing the opportunities for corruption. In addition, the EU has initiated a large-scale transition from paper to electronic tendering procedures over the past decade. E-procurement platforms such as Prozorro in Ukraine and TED (Tenders Electronic Daily) in the EU provide a full "electronic trail", increasing transparency and accessibility of information for all stakeholders.
The platform that operates in Ukraine provides a unified information space where tender announcements, commercial offers, and contracts are published. At the same time, in Ukraine, there have been significant problems with access to information since the beginning of the full-scale invasion. Limited access to legal registers prevents full monitoring and timely detection of corruption risks. This circumstance significantly reduces the effectiveness of the state's anti-corruption policy.
The digitization of legal algorithms as a tool for mitigating corruption in public procurement is
explored [
The use of quantitative indicators to detect corruption risks in public procurement is
researched [
The object of the research is the processes of public procurement and tendering within Ukrainian
electronic registry systems. The subject of the research is the methods and digital tools for analyzing
data from state registries to identify anomalous patterns and detect corruption risks in tenders and
procurement activities. To reduce the corruption risks in Ukraine, the operational control over the
transparency of procurement is a relevant problem [
One of the common methods of abuse is to create the appearance of competition when related companies participate in a tender. This can happen through several mechanisms: • affiliation by managers and beneficiaries (several legal entities belong to the same group of owners); • using registration of fictitious firms to submit alternative bids and simulate competition; • price manipulation (preliminary agreements between participants on the distribution of winnings by overstating or understating offers).
The proposed web application automatically collects data via API on the following information: about tenders, amounts of concluded contracts and data on contractors, subcontractors, including their identification code (EDRPOU), legal addresses, information on directors and ultimate beneficiaries. After collection, the data is consolidated in a single database.
The analysis is carried out according to the following key parameters: • intersection of beneficiaries, directors and legal addresses of companies; • frequency of victories of one company in tenders from one customer; • relationship of participants through third structures (affiliation of companies); • unusual changes in the cost of tender offers.
It helps to identify companies that systematically win tenders from the same subcontractors, and find the connections between the winners of purchases through founders, directors and registration addresses.
The program implements the analysis of tender processes, including the selection of the contractors, viewing detailed information on purchases and analyzing the winners. The system displays data on tenders, including the name of the purchase, the winner, the contract amount and the list of participants. Information on the winner is interactive, which allows for immediate in-depth analysis (Figure 1).
This tool is effective for monitoring public procurement, helping journalists and anticorruption agencies quickly identify possible violations. By using open data and modern analysis algorithms, the web application promotes transparency and fairness of procurement processes.
Based on the analysis of several tenders by the system, the characteristic patterns indicating possible corruption risks are proposed and represented in Tables 1 and 2. Analysing the distribution of financial flows allows not only to determine the share of the budget - Total amount of tenders; attributable to each contractor, but - Average contract price; also to identify anomalies - The share of the budget going to systematic and disproportionate one subcontractor allocation of funds to one bidder. - Average difference between the Such behaviour may indicate possible forecast and actual contract price affiliation with the contracting authority, restriction of competition or the existence of corrupt agreements.
Financial flows
interested in some specific subcontractors based on personal gains, but on professional requirements
- Distribution of tenders by month (are there any spikes in During wartime, a minimum of 7 November/December?) days is allowed for conducting a - The share of tenders conducted tender, usually, it is 15, and if it is urgently (less than 10 days between often 7, then it is suspicious; it may the announcement and the be a tender for one person. submission of applications)
Analysis of the temporal characteristics of participation in tenders allows us to identify typical and anomalous patterns of contractor behaviour - for example, a concentration of wins in certain periods (usually at the end of the budget year), bursts of activity in November-December, and irregular or selective frequency of participation. Such temporal anomalies may indicate attempts to target budget allocation in favour of a particular bidder, participation in pre-agreed procedures, or fictitious competition.
To further enhance the system for tender analysis, we can utilize a machine -learning-based approach. The cluster analysis is used to identify groups of companies exhibiting anomalous behavior in tenders. This approach enables the segmentation of companies into clusters based on a multidimensional feature space, such as encompassing tender activity, behavioral patterns, affiliations, geography, interactions with contracting authorities, and financial indicators. Additionally, it aids in identifying companies that significantly deviate from typical behavioral profiles. By utilizing unsupervised machine learning techniques such as k-Means clustering, DBSCAN, and Isolation Forest, it is possible to detect individual companies or small groups exhibiting unusual patterns. To do so, first, we need a proper dataset to train our model with.
There are many types of datasets based on the type of data they contain. In our case, we have a tabular dataset, which is a database dump into CSV.
The dataset consists of 3 CSV files, each represents a specific table in the database schema. Contractors.csv file represents information about the companies that put tenders, as shown in Table 3.
The file tenders.csv represents information about tenders that are submitted by the companies, as shown in Table 4.
The file subcontractors.csv represents information about the customers who win/participate in those tenders, as shown in Table 5.
The current dataset schema is not complete since right now some of the government data sources are closed due to the war. Because of this, we can only use those that are in the public domain. The data in the public domain is quite good, however, in some cases, the information is incomplete or not enough for analysis.
In particular, the following categories of information are missing: • history of changes in founders, , and ultimate beneficiaries, which is critical for identifying hidden company affiliations; • data on the company's turnover, debts, and VAT payments, which is necessary to separate real businesses from fictitious firms; • data from declarations of civil servants is important for establishing connections between officials and companies that win tenders; • data on the execution of contracts - including the involvement of subcontractors, certificates of completion of work, and payment, which is often not published; • historical tender data, including for defense procurement and critical infrastructure facilities, which are currently closed. contractor winner_edrpou winner amount date subcontractors created_at url_tenders
The absence of these data limits the capabilities of machine analysis and does not allow for the full identification of affiliation schemes, price gouging, fictitious competition and other signs of corruption. Gaining access to such registers, at least in a limited form for scientific and analytical purposes, is critically important for building an effective public procurement monitoring system in Ukraine.
Another difficult in tender data analysis is the different terms in text type attributes may represent the same concept. Appling Natural Language Processing technologies during the preliminary data preparation phase to identify synonyms in text data is an effective approach that enhances the quality of subsequent analyses.
Thus, the use of modern technologies in public procurement is an important step in the fight against corruption. Digital platforms improve the transparency of procedures and provide information for analysis and identification of possible corruption schemes.
Using web applications for monitoring tenders helps to effectively identify fictitious competition schemes and contract manipulation. The system for identifying relationships between companies helps to minimize the influence of affiliated structures on procurement results.
However, significant challenges remain, such as insufficient access to data, the complexity of integrating different sources of information and the limited capabilities of law enforcement agencies. To achieve maximum effect, it is necessary to further develop the legislative framework, ensure open access to state registers and improve data analysis methods.
To improve the accuracy of analysis and further automated detection of corruption schemes, we propose the formation of a dataset that will include: • historical data on tenders and their winners; • information on company connections, directors and beneficiaries; • financial indicators and amounts of concluded contracts; • tags of previously identified corruption cases and investigations; • data on unnatural price jumps and discrepancies with market conditions.
This dataset will be used to train machine learning models that can automatically find suspicious anomalies and offer analytical reports for experts and government agencies.
This direction of the system's development is aimed at use by both government anti-corruption agencies and independent researchers and journalists.