Business intelligence, statistical audit, Gaussian mixtures
A Business Intelligence (BI) System employs tools from several areas of knowledge for the collection, integration and analysis of data to improve business decision making. The Brazilian Ministry of Planning, Budget and Management (MP) uses a BI System designed with the University of Bras´ılia to ascertain irregularities on the payroll of the Brazilian federal government, performing audit trails on selected items and fields of the payroll database. This current auditing approach is entirely deterministic, since the audit trails look for previously known signatures of irregularities which are composed by means of an ontological method used to represent auditors concept maps. In this work, we propose to incorporate a statistical filter in this existing BI system in order to increase its performance in terms of processing speed and overall system responsiveness. The proposed statistical filter is based on a generative Gaussian Mixture Model (GMM) whose goal is to provide a complete stochastic model of the process, specially the latent probability density function of the generative mixture, and use that model to filter the most probable payrolls. Inserting this statistical filter as a pre-processing stage preceding the deterministic auditing showed to be effective in reducing the amount of data to be analyzed by the audit trails, despite the penalty fee intrinsically associated with stochastic models due to the false negative outcomes that are not further processed. In our approach, gains obtained with the proposed pre-processing stage overcome impacts from false negative outcomes.
Pilon, Bruno Hernandes Azenha; da Costa, Joao Paulo Carvalho Lustosa; Murillo-Fuentes, Juan J.; and de Sousa, Rafael T. Júnior, "Statistical Audit via Gaussian Mixture Models in Business Intelligence Systems" (2015). Proceedings of the XI Brazilian Symposium on Information Systems (SBSI 2015). 10.