Ontologias difusas no suporte à mineração de dados: aplicações na Secretaria de Finanças da Prefeitura Municipal de Belo Horizonte

Abstract

This research project aims at the improvement of technology support for detecting, analyzing and fighting tax evasion in Service of Any Kind Tax- ISSQN - from recovery and processing large volumes of data. The hypothesis is that the association of methodologies and techniques of ontologies and fuzzy systems, can even help facilitate the success of data mining in the recovery of these large volumes of data. This hypothesis is based on three premises. The first premise is that the solution uses the capture, processing, modeling, representation and formal incorporation of domain knowledge, in particular, constituted by thecollective knowledge of experts. The second premise is that ontologies are perfectlysuited to the task of knowledge incorporation. This adequacy is obtained in view ofthe characteristics of ontologies in explicit, formalize, verify and consolidate the knowledge, making them sharable, reusable and interoperable. In addition, the knowledge represented and incorporated is naturally used for classification, from the inherent inference mechanisms of development tools ontologies, and the mechanisms ofinferences that can be added to them. The third premise is that, considering that the collective knowledge of experts is vague and subjective in nature, methodologies and techniques in the area of fuzzy systems are adequate to capture, treat andmodel this knowledge. Then, we have developed a solution for the pre-and post-processing of data mining, focused on modeling expert knowledge of nature vague and subjective. This vague and subjective knowledge is modeled to attributes with fuzzy systems techniques, guiding the process of data mining, and generating a subjective measure that supports the analysis and interpretation of results that otherwise would be more laborious, difficult or even impossible to be performed. Thus, we present an effective solution for increasing tax revenue through the identification of evidence of fraud and tax evasion on imported services for companies of Belo Horizonte. Based on this approach, tests and simulations conducted, we can reduce the numberof rules of interest generated by data mining by 42%. Comparing the number of records in data mining detected by this approach, with the number of known records involving fraud proven, we obtained an overall success rate of 95.88%. This solution has the potential to be applied in other situations, and in wide areas of application in the public and private sphere. The project explores the convergence of skills developed by three different institutions: the School of Information Science at UFMG, the Department of Computer Science at UFMG and the Finance City Office of BeloHorizonte.