Data mining for fraud detection using invoicing data. A case study in fiscal residence fraud
This paper describes a methodology to eciently build predictive fraud detection models based on payment transaction data. More specifically, a network learning technique is applied using invoicing data from and to foreign companies. A network is created among foreign companies, where two companies are connected if they have sent an invoice to (or received an invoice from) the same Belgian company. These connections are weighted, taking into account the number of shared Belgian companies and the popularity of the Belgian company that links the foreign companies. Data mining techniques are applied to predict residence fraud committed by foreign companies. Our empirical results show that the obtained models are indeed able to discriminate between fraudulent and non-fraudulent companies, with an AUC up to 79%. The superiority of our proposed method is shown by comparing its results to a support vector machine trained on the same transacitonal data (including SVD and balancing of the dataset).
Year of publication: |
2013-10
|
---|---|
Authors: | MARTENS, David ; JUNQUÉ DE FORTUNY, Enric ; STANKOVA, Marija |
Institutions: | Faculteit Toegepaste Economische Wetenschappen, Universiteit Antwerpen |
Saved in:
Saved in favorites
Similar items by person
-
Data mining for fraud detection using invoicing data : a case study in fiscal residence fraud
Martens, David, (2013)
-
Classification over bipartite graphs through projection
STANKOVA, Marija, (2015)
-
Active learning-based pedagogical rule extraction
JUNQUÉ DE FORTUNY, Enric, (2014)
- More ...