The rapid development of technology has digitized customer payment behavior towards a cashless society. To a certain extent, this has created a feast for miscreants to commit fraud. According to Nilson (2020), global fraud loss is projected to reach over $35 billion by 2025. Consequently, the need for a novel method to prevent this menace is undisputed. This research was conducted on the IEEE-CIS Fraud Detection Dataset provided by Vesta Corporation. Based on the logic of labeling for converting the entire account to ''Fraud=1'' once the credit card has fraud, we navigate the research process towards predicting fraudulent credit cards rather than fraudulent transactions. The key idea behind the proposed model is user separation, in which we divide users into old and new people before applying CatBoost and Deep Neural Network to each category, respectively. In addition, a variety of techniques to improve detection accuracy, namely handling heavily imbalanced datasets, feature transformation, and feature engineering, are also presented in detail in this paper. The experimental results showed that our model performed well, as we obtained AUC scores of 0.97 (CatBoost) and 0.84 (Deep Neural Network).
Rapid development of technology has digitized customers' payment behavior towards a cashless society. To some extent, this has created a feast for miscreants committing fraud. According to Nilson’s Report (2020), the global fraud loss is projected to reach over $35 billion in 2025. Thus, the need for a novel method to prevent this menace is undisputed. Studies in fraud detection in cards abound with various machine learning-based techniques. However, many of them were conducted on modest-sized sample with a few familiar features and mainly focused on transaction-based approach to detect fraud. These models fail to satisfy low false alarm rates when being fed with large-scale data streams. In this paper, an innovative model was proposed to improve classification accuracy based on the idea of user separation, in which we divided users into old and new people before applying CatBoost and Deep Neural Network on each category, respectively. The experimental result shows that our model achieved better as we obtained AUC score of 0.97 (CatBoost) and 0.84 (Deep Neural Network).
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
customersupport@researchsolutions.com
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.