Public Administration Suppliers Classification Model Based on Supervised Machine Learning

Abstract

Context: Public contracts are agreements made between the Public Administration and individuals to achieve public interest objectives. Within this relationship, some problems such as contractual breaches may occur. In the Piauí State, the Audit Court, environment in which this research was developed, is responsible for analyzing and judging the accountability of the Legislative, Executive and Judiciary Powers.. Problem: For government control bodies, the challenge generated is to act efficiently in the identification of fraud and corruption. To audit all processes, there is an unfeasible number of records to be analyzed by each auditor. Solution: Optimize the choice of processes to be audited, given the infeasibility of a full census. Thus, the present work uses Machine Learning (ML) techniques to assist in the selection of which ones will be audited. IS theory: Machine learning studies the computational methods that allow computer programs to autonomously obtain an improvement in a given task through experiments. Method: After the preparation applying the balancing and normalization of the base provided by the Audit Court that gathers other datasets about suppliers, experiments were carried out and the J48 algorithm was identified as the most appropriate for classification through the decision tree structure. Summary of Results: The constructed model resulted in a correct classification rate above 82% to solve the problem of classifying suppliers as high and/or low risk. Contributions and Impact in the IS area: The resulting classification model is expected to serve as support for an automatic supplier evaluation and classification system.