Detecção de anomalias em dados da administração pública utilizando técnica de aprendizado de máquina

Abstract

The use of Machine Learning (ML) has been used in several areas of society and organizations. With this several applications and discoveries have arisen from this area. The objective of this work is to find data anomalies related to expenditures provided by municipalities using techniques of machine learning and data mining. For the development of this work, the Python programming language, libraries, Pandas, NumPy, SciKit-learn is used. The first step was to select the data that have some type of correlation, so we chose to use the data from the Accounting and Fiscal Information System (SICONFI) of the National Treasury portal, more specifically data on municipalities’ spending on education , health, and public safety, among other attributes. They were then unified with data from the Firjan Index of Fiscal Management (IFGF), Firjan Municipal Development Index (IFDM), data from the Department of Information Technology of the Unified Health System (DataSUS). With this new proposal, new anomalies of interest were found. Other cities that, in the experiment with only the SICONFI base, were considered anomalies in this new approach, except in the ten cases analyzed, the expenditure of these cities was justified by the proposed indexes.