TOWARDS SAFER BANKING: MODELING TRANSACTIONS WITH GENETIC INSIGHTS

Abstract

Credit card fraud poses a significant financial threat, with the United States incurring annual losses of approximately $190 billion. Harnessing artificial intelligence (AI) holds immense potential for substantially mitigating these losses. This potential lies in the ability of banks and credit card companies to employ sophisticated methods for predicting and flagging fraudulent transactions, akin to identifying "bad genes." This identification relies on an amalgamation of account data, event specifics, geographic locations, and prior transactions marked as fraudulent. In essence, these elements form the unique "DNA" of transactions (Viaene et al., 2002). This study endeavors to delve into the intricacies of genes, their constituent components, and the fundamental architecture of genes, drawing parallels with transaction modeling. Additionally, this research embarks on a personal exploration of my banking history, dissecting transaction attributes that could have prompted the bank to flag certain activities as potentially fraudulent, enabling preemptive intervention before funds are credited to my account. To facilitate this investigation, two datasets containing transaction information will be leveraged for rigorous analysis. One dataset, dated 2013, offers pertinent but less detailed information, while the other, originating from 2004, presents a comprehensive array of transaction attributes. Both datasets encompass transactions marked as fraudulent and unmarked ones. The earlier dataset encompasses particulars such as transaction type, amount, associated company, and chronological data. Hyperlinks to this dataset are meticulously outlined in the data description and the supplementary appendix (Shaughnessy, 2011; Datasets; Kou et al., 2004; Dasarathy, 2009).