Detecção de anomalias em sistemas de administração de frotas públicas municipais

Abstract

Endemic corruption is difficult to detect and treat, being a solid link of impunity. In computeri- zed systems, which are part of the complex public management network, it is possible that legitimate data are manipulated with an unlawful intention, shadowing reality with the purpose of evading the tax authorities. Detecting such patterns of illegitimacy is not a trivial task and, in general, it is associated with interpretive aspects of legitimate data before the system. As this interpretation depends on tracking and crossing a wide and complex tangle of information, of multiple natures, such as the web, external bases, state and federal public interfaces, etc., this task is often unfeasible, becoming the backbone of endemic corruption. In this article, data science techniques are applied to support the extraction of irregularity patterns and trends in real public databases. In particular, the LOF algorithm is applied, based on the concept of local density through k nearest neighbors, where distance is used to assess density, on public databases relating to fleet maintenance, to identify potentially transactions fraudulent. Two analyzes were carried out to illustrate the method, one concerning the refueling of vehicles with gasoline and the other concerning the refueling of vehicles with diesel oil. Results show that the performance of the algorithm evidences the existence of refueling records for vehicles above the fuel tank capacity, with 27 for refueling with gasoline and 90 for refueling with diesel oil.