Abstract

Detecting collusion in public procurement is critical to ensure fair and transparent practices in government acquisitions. Bid collusion in auctions poses a major challenge in public procurement by causing unfair price hikes through unlawful cooperation among competing firms, consistently affecting the overall supply chain. This study uses machine learning methods to investigate collusion in public procurement processes. It delves deeply into exploring multiple machine learning models such as random forests, extra tree classifiers, support vector classifiers, Neural Networks, Gradient Boosting, and various combinations of models for collusion detection. First, the models were trained using available data, followed by the inclusion of screening variables derived from bid information as additional features. The additional features were fed to the models, which went through fine-tuning of parameters. Additionally, comparative analyses were carried out to evaluate the merits and drawbacks of each model. Metrics including Accuracy, balanced accuracy, precision, recall, F1-score, and ROC-AUC score were evaluated, providing a comprehensive evaluation framework. Various settings were used to compare which set of inputs gives the highest accuracy in collusion detection. The ROC-AUC analysis brought forward crucial insights, particularly regarding models' abilities to minimize false positives while maximizing true positives. Models like Random Forest and Gradient Boosting demonstrated superior performance, showcasing lower false positive rates—a crucial aspect when identifying collusion in public procurement. Additionally, the study underscores the significance of feature engineering in collusion detection. Specifically, attributes like screens - CV, SPD, DIFFP, RD, SKEW, KURTO, and KS significantly aid algorithms in processing data effectively to identify collusion patterns. The outcomes of this study carry significant implications for both the specific domain under investigation and the broader field of collusion detection. Ultimately, this research provides a valuable guide for policymakers, procurement officers, and data scientists, offering valuable insights into the effective machine learning techniques tailored for detecting collusion in public procurement.