Article Preview
TopIntroduction And Problematic
With the great of the global economic entity from dealing in cash to credit card transactions, which are a plastic card carrying a set of information such as full name, account number, expiration date and company logo (Zareapoor & Shamsolmoalia, 2015), the card also holds an electronic chip that allows it to be used in electronic payment devices, and can be recharged with money and used for direct purchase or online and has become more common in everyday life, (Jurgovsky, Granitzer, Ziegler & Calabretto, 2018) as shown in Figure 1.
However, the use of these cards is not without risks. Illegal transactions may occur at the expense of one person by another person making purchases or transfers without the knowledge of the cardholder by stealing one or more credit card information, what is considered to be theft and infringement of privacy, and therefore developed several systems to detect these frauds to reduce losses of dealers.(Siddhartha, Sanjeev, Kurian & Christopher, 2011) Fraud rates soared dramatically as online sales grew where the UK card fraud totaled £ 671.4 million in 2018, or 19 percent, An increase from £ 565.4 million in 2017.At the same time, total spending on all debit and credit cards amounted to £ 800 billion in 2018, with 20.4 billion transactions made during the year.(Anon, 2018) What distinguishes these scammers is the constantly changing strategy and the low percentage of fraud compared to the original transactions, which make it difficult for systems and investigators to detect fraudulent transactions, this have led to the adoption of these systems on learning algorithms, which in turn suffered from the difficulty of retaining most of the assumptions in the fraud detection system, which is a lack of realism. Credit card fraud detection is a fairly mature area of research, prompting researchers try to develop fraud detection systems to gain access to the system capable of accurately detecting unwanted transactions. Most of the previous work in this field relied on traditional machine learning classifiers that rely on features manually extracted from training data. (Fang & Boli, 2021) (Shukur & Kurnaz, 2019) In this paper, one of the most advanced techniques in the field of classification called deep learning has been proposed, generally designed in the form of two external layers (input layer and output layer) within which one or several hidden layers, and the recurrent neural network RNN is one of the most effective deep learning techniques, (Vries & Principe,1991) The research will be modeled as a new enhanced prototype named Long Short Term Memory (LSTM) which can successfully identify fake credit cards. What distinguishes LSTM from the normal neural network is the presence of additional subjective loops that help retain the wrong information as well as pass it to the next layer with a large number of hidden layers or processing layers. The LSTM data retention feature gives it the ability to choose which information to store, which should be ignored and which should be transferred to the adjacent layer. Unlike ordinary neural networks that lack this characteristic as each layer is unique treatment without resorting to the adjacent layers. (Victor & Muthu, 2015)
The biggest problem facing credit card fraud is the speed of response with accuracy in the decision to quickly understand any problem or financial embezzlement. His application of this technique was not an easy matter, as due to the difficulty of adjusting the parameters and the nature of the data set, the details will be mentioned at the end of this paper.
The proposed solution in this paper is a very effective solution with the ability of the proposed algorithm to remember the sequence of all transactions and judge any new transaction as being accepted or rejected.