Classification of public procurement procedures based on red flags with a machine learning approach for the detection of possible risk of corruption

Abstract

This study addresses corruption in public procurement by integrating machine learning and red flags. Objectives include evaluating red flags’ impact on corruption detection using machine learning models and comparing different algorithms’ performance. Using data from Mexico’s CompraNet platform, red flags—indicators of potential corruption— are incorporated. Supervised machine learning models (e.g., XGBoost, Random Forest, Logistic Regression) are trained and evaluated with using both inputs, with and without red flags variables. The key findings of this study underscore the significant positive influence of red flags on the accuracy of corruption detection across various machine learning models. The integration of red flags consistently improves precision, recall, and F1-scores, reaffirming their effectiveness as valuable corruption risk indicators. Furthermore, the comparative assessment of machine learning algorithms reveals variations in performance, emphasizing the critical nature of model selection. In conclusion, red flags effectively help to improve the detection of potential corruption risks in public procurement. Machine learning’s role in leveraging red flags shows promise for corruption detection. These insights have implications for public governance and policymaking, emphasizing the potential of data-driven approaches in mitigating corruption’s adverse effects. Furthermore, this research highlights avenues for future exploration, such as tailored red flag frameworks, real-time detection, and cross-domain application, providing a comprehensive outlook for advancing corruption detection and prevention strategies.