Article Preview
Top1. Introduction
Today, machine learning algorithms have become more powerful and efficient for solving more complex problems in our world, which is due to the huge amount of data produced by humanities and the advancement in technologies and machine capacity. Credit card fraud is one of those complex problems today. Many banks and financial institutions have suffered from the number of fraud transactions committed every year (Kim et al., 2019). A credit card is a physical medium used by a bank or a credit union to enable cardholders to pay a merchant for goods and services online or outline (Awoyemi et al., 2017). Credit card fraud detection is the procedure to detect fraudulent transactions. Credit card crimes occur when the card identities such as credit card number, expiration date, password, and name of the credit cardholder (Thennakoon et al., 2019), are stolen by the fraudsters to make fraudulent transactions. The operation of payment can be described as follows; first, the cardholder uses their credit card to pay for goods or services by presenting their credit card to a merchant, after that the merchant uses their credit machine software/gateway to send a request to their acquiring bank or the bank processor containing the cardholder’s information. The acquiring bank or its processor captures the transaction information and routes it through the adapted card network Vista or MasterCard networks to decide if the transaction is authorized or not based on the transaction information (Hossain & Uddin, 2018).
Many techniques are proposed to catch fraudulent transactions in credit cards; the recent one used machine learning algorithms as a solution. These algorithms show their strength ability (Tayebi & El Kafhali, 2023) and efficiency in distinguishing between fraudulent and legitimate transactions (Caroline Cynthia & Thomas George, 2021). There are two mean approaches for fraud transaction detection; the first one uses supervised learning algorithms (Khatri et al., 2020), which need the dataset to be labeled, and the algorithm uses this label to learn patterns that exist to separate fraud from non-fraud transactions. As an example of the used supervised learning methods we found, the neural networks (Esenogho et al., 2022), regression model (Hussein et al, 2021), Naive Bayes (Gupta et al., 2021), decision trees (Bahnsen et al., 2019), Fuzzy Logic (Razooqi et al., 2016), genetic algorithm (Tayebi & El Kafhali, 2021), and particle swarm optimization (Tayebi & El Kafhali, 2022). The second approach uses unsupervised learning algorithms to detect fraudulent transactions (Carcillo et al., 2021); this method aims at using unlabeled data to group samples into fraudulent and legitimate transactions. As an example of this approach, we found in the literature Self-Organizing Maps (SOMs) (Olszewski, 2014), K-Means (Rai & Dwivedi, 2020).