Using Bayesian networks to identify and prevent split purchases in Brazil
Rommel Novaes Carvalho, Leonardo J. Sales, Henrique A. Da Rocha, Gilson Libório Mendes · 2014
To cope with society’s demand for transparency and corruption prevention, the Brazilian Office of the Comptroller General (CGU) has carried out a number of actions, including: awareness cam-paigns aimed at the private sector; campaigns to educate the public; research initiatives; and regular inspections and audits of municipalities and states. Although CGU has collected infor-mation from various different sources – Rev-enue Agency, Federal Police, and others –, going through all the data in order to find suspicious transactions has proven to be really challenging. In this paper, we present a Data Mining study ap-plied on real data – government purchases – for finding transactions that might become irregular before they are considered as such in order to act proactively. Moreover, we compare the perfor-mance of various Bayesian Network (BN) learn-ing algorithms with different parameters in order to fine tune the learned models and improve their performance. The best result was obtained us-ing the Tree Augmented Network (TAN) algo-rithm and oversampling the minority class in or-der to balance the data set. Using a 10-fold cross-validation, the model correctly classified all split purchases, it obtained a ROC area of.999, and its accuracy was 99.197%. 1