PREDICTIVE MODELING FOR THE DETECTION OF ANOMALIES IN PUBLIC CONTRACTS

Abstract

Public procurement represents one of the most significant areas of government expenditure and, consequently, one of the most vulnerable to inefficiencies and irregularities. This study proposes and validates a predictive data-science framework designed to detect anomalies in public contracts, specifically focusing on the probability of financial amendments (aditivos) as indicators of potential deviations. The research follows an end-to-end methodological structure, encompassing data acquisition from open-government sources, preprocessing, feature engineering, and the implementation of a Gradient Boosting Machine (GBM) model optimized for highly imbalanced datasets. Empirical validation revealed strong performance, with the model achieving a recall rate of 0.85, emphasizing sensitivity over precision to minimize the non-detection of real irregularities. Beyond technical development, the study also discusses the necessity of Explainable Artificial Intelligence (XAI) for algorithmic transparency and explores the Blockchain technology as a potential foundation for next-generation auditing ecosystems. Ultimately, the paper contributes a reproducible roadmap for algorithmic governance, strengthening proactive oversight mechanisms and supporting data-driven decision-making in the public sector.

DOI: https://doi.org/10.56238/edimpacto2025.065-001