International Journal of Engineering Research and Reviews
ISSN 2348-697X (Online) Vol. 10, Issue 4, pp: (1-8), Month: October - December 2022, Available at: www.researchpublish.com
DEBIT AND CREDIT CARD FRAUD DETECTION USING K-N-N IN MACHINE LEARNING Ahmad Alammar1, Yazeed Al Moayed2, Nasir Ahmed Algeelani3 Faculty of computer science and Information technology AL-Madinah International University Kuala Lumpur, Malaysia DOI: https://doi.org/ 10.5281/zenodo.7260970
Published Date: 28-October-2022
Abstract: We have thousands of banks around the world and these banks revenue through a very small profit from each transaction. So from the bank perspective, they have no idea who is performing these transactions whether they are trusted or untrusted. All they know is that the transaction details have a correct card number and CVV and the user has input the transaction details successfully. but they still lack the ability to detect whether these transactions are coming from an authorized person or not. So whenever a fraud incident happens then the data analysis team needs to study the data and investigate the data and decide whether the transaction was trusted or not. If it was a fraudulent transaction, then the bank needs to compensate the user. This paper will discuss several fraud detection mechanisms that exist in the market already and will propose a new machine learning mechanism that helps to detect fraudulent transactions and helps industries to mitigate the fraud incident. Keywords: K-N-N algorithm, machine learning, Fraud Detection.
I. INTRODUCTION The card payment business generates revenue through volume very large number of transactions with little profit margin per transaction. Fraud is rare (on the order of 1 fraud per thousand transactions, depending on the context), this can significantly reduce the margin profit of the financial institution since the refund of the amount of a single transaction can be equal to the sum of the marginal profit of a large number of legitimate transactions. That is why they exist in the market different computer tools that can study each customer's transactions and, through some mechanism, produce an alarm when any of them seem suspicious [1]. For example, these could be alerted by employing an application to the cardholder about the alarm generated, calling him to verify the integrity of the purchase, or eventually blocking its plastic. However, the management of alerts has a direct cost associated (salary of analysts, cost of calls, etc.) and indirect, but also relevant (inconvenience to clients, loss of trust in the institution), which can be a potential source of losses for the entity. It creates tension between the extreme cases of not generating too many alarms (then there is no detection of almost no fraud) and avoiding most fraud (at a cost, possibly unsustainable analysis) [2]. The search for balance between these two extremes is a non-trivial and extremely important problem: spending more and more time and resources. That poses an interesting challenge at the scientific level to develop new and better detection models. Also, in recent years, as there are advancements in technology, and most of them are using credit or debit cards for buying their needs and buying online products, the fraud associated with it is also rising gradually and the risk of stealing card information in increasing especially when the user inputs his card details to unknown website to try to purchase items online.
Page | 1 Research Publish Journals