Detection of debit card fraud through random forest

As one of the most frequently used financial tools in our life, ATMs have become a target for fraudsters in the same frequency. Particularly, security vulnerabilities of debit cards, which are generally produced by using magnetic stripes, was seen as an opportunity for fraud. As a result of exploiting those security vulnerabilities, important amounts have been fraudulently withdrawn from customer accounts. In this thesis, a data mining model was established for detection of debit card fraud through debit card transaction data of a bank. Firstly, transaction variables were defined in the ATM cash withdrawal dataset with consideration of their relevance in the debit card fraud detection. Consequently, behavioral RFM (Recency, Frequency, Monetary) variables, which are suggested as relevant in debit card fraud detection literature, were calculated based on those transaction variables. Secondly, several experiments were made through the classification model created by random forest algorithm by changing algorithm parameters. In the concluding remarks, the results of the established model were summarized and, considering practical implementations, some assessments regarding a real-time debit card fraud detection system were made.