Bayesian Dialysis of the Evidence in Fraud Detection

Abstract

In this paper we propose a methodology that combines Bayesian and “big data” tools designed to optimize the investigation of fraud that we call “Bayesian dialysis”. Its use improves the results of any statistical selection technique previously used, capitalizing on the evidence obtained by decision makers. Its effectiveness is shown using as an exemplary case the selection for VAT control in the “Agencia Estatal de Administracion Tributaria” (AEAT), that is the State Tax Agency. Although in Spain the existing systems for the selection of taxpayers in the control of fraud in VAT, in a context in which the impact of the fraud is estimated at 2%2, get excellent results, the proposed methodology, which has the characteristics of novelty and generality, improves accuracy in the detection of fraudsters in a 12.29%, from an average of 82.28% fraudsters detected in the collectives inspected to 94.36%. In addition, the confidence interval in the detection of fraud, at the 0.95 level, improves by 13.72%, which represents a qualitative leap in the effectiveness of the control reaching the threshold of 95.51%