Article Preview
Top1. Introduction
With the exponential growth of technology, the internet, and the electronic market, credit card prediction has become an exciting research topic. Online transactions also play a vital role in the present scenario. Electronic commerce also employs Big data and artificial intelligence techniques to gain profits in their business (Maes et al., 2002; Niu et al., 2019). While most transactions and companies are running online, payment transactions are also online and generate massive data. Online payments are made using credit cards, debit cards, net banking, and UPI options. The online credit card payment system needs a predictive model to distinguish whether a transaction is a fraud or not (Maes et al., 2002). Although numerous modern prediction models are proposed to detect credit card fraud transactions, accuracy is still challenging (Niu et al., 2019; Odegua, 2020). The existing model lacks inefficiency because of the hugeness of transaction data and data imbalance problems. Imbalanced data refers to whether one class has more instances than another class category. It leads to the class imbalance problem (Divakar & Chitharanjan, 2019; Zhu et al., 2017). Another Challenge is a significant transaction because a large data set has increased heavy-tailed noise distribution and nonlinear patterns. Therefore, traditional approaches are not able to gain higher accuracy (Petropoulos et al., 2019). The preliminary study found that supervised and unsupervised learning techniques have been used to uncover credit card fraud and forecasting (Randhawa et al., 2018). Chan et al. recommended a cost model for fraud detection (Chan et al., 1999). They have combined multiple fraud detectors in the cost model and demonstrated that the loss had been reduced due to distributed data mining of fraud models. A fraud detection model is suggested by Bolton et al. using user behavior for a credit card transaction (Bolton & Hand, 2001). Zojaji et al. review various techniques, data set, and evaluation criteria for credit card fraud transactions (Zojaji et al., 2016).
Figure 1.
Process of credit card fraud discovery
Figure 1 shows the process of credit card fraud discovery. This process described those various essential conditions that are checked before completing the transaction. It includes enough balance, card number, expiry valid, PIN details, etc. According to the provided details prediction model, it classifies transactions as fraud or genuine. This paper investigated the analysis of credit card fraud prediction using Gradient Boosting algorithms XGBoost (Chen & Guestrin, 2016). In addition, we perform an extensive comparison of the prediction accuracy of the XGBoost approach with other machine learning approaches. We also presented evaluation matrices, such as accuracy, precision, recall, and F1 score.
Further, the paper is organized into four sections: Section 2 provides a relevant overview of credit card fraud discovery. Section 3 explains the selected models. In section 4, we have discussed our outcomes and discoveries, and section 5 concludes.