Integer Formulation for Computing Transaction Aggregation to Detect Credit Card Fraud
Résumé
Banks assume great costs caused by fraudulent credit card transactions. With the development of new means of payment and the virtualization of banking services, fraud is becoming more and more difficult to detect and its frequency is growing at a dizzying rate.
Currently available defense tools are not enough to counter fraud in continuous evolution. In this work, we present an integer formulation to compute functions of transactions data that turn to be useful for detecting behavioral patterns present in past fraudulent events. Our formulation allows the aggregation of transactions over any set of features, in addition to filtering transactions that verify structured requirements. We test our model with realworld data from a French bank, using SAT solvers. Numerical results obtained on several instances show the effectiveness of our approach.