Abstract
Internet-based payment methods in recent times are rapidly growing both in developing and developed economies. Credit card-based payment systems are among the prominent cashless payment methods in all economies. Credit card frauds by cyber-criminals are increasing in spite of several precautionary measures. Thus, fraud detection in real-time is a challenging task. Several machine learning tasks have attempted to solve the problem. This chapter proposes a two-step method to detect credit card fraud by coupling the deep learning-machine learning approaches. In the first stage, the dimensionality of the data set is reduced to 50% by a deep auto-encoder. A machine learning classifier classifies the instances in the second stage. Among the machine learning algorithms, the CatBoost and Random Forest achieved better performance. Their performance aligned with the state-of-the-art approaches. The proposed method is robust against the labor-intensive feature selection and imbalanced class problems.
TopIntroduction
Internet-based payment methods in recent times are growing at a rapid pace in developed and developing economies. The credit card industry players are losing billions of dollars despite an adequate level of security measures. Fear of losing the public and customer trust, credit card firms are typically reluctant to declare such information. Therefore it is difficult to approximate the losses incisively. This kind of information hiding had resulted in the incapability to precisely audit such losses. From 2010 to 2018, the number of fraud losses per $100 of card transactions is shown in Figure 1 (Szmigiera, 2021). The fraud loss forecasts for 2027 are also documented. The leading global card and mobile payments trade publication Nilson Reports (Nilson Report, 2019) published that 27.85 billion U.S. dollars were lost due to frauds worldwide in 2018. Accordingly, the loss is projected to increase to 35.67 billion U.S. dollars in next five years and 40.63 billion U.S. dollars in ten years.
Most researchers have attempted to use classification or clustering methods in the automated detection of such frauds. In reality, the number of such fraud transactions is lesser than the original or regular transactions. Thus, this skewed nature of the data set made the detection of credit card fraud transactions a more meaningful and exciting problem to be solved. Detection of these small numbers of fraudulent card transactions would save a considerable amount of money for the credit card firm and the credit cardholder. Thus, there is an increasing research trend on detection methods for credit card fraud detection. Application (Phua, Gayler, Lee, & Smith-Miles, 2009) and behavior frauds (Bolton & Hand, 2002) are the most commonly found credit card frauds. Application fraud is carried out by using Identity theft. In this type of fraud, the fraudster applies to the issuer for a credit card with false documents. The banks take several protection actions in curbing such occurrences. It includes knowing the customer (KYC) document verification, calling the employer frequently, and so on. Behavioral fraud happens when the correctly issued credit card is fraudulently used in a transaction.
The proposed approach uses a two-step approach that employs a deep autoencoder (DAE) and a classifier. This approach uses a reduced size input vector to classify the type of transaction as genuine or fake. The main contribution of the proposed work is the use of the Deep-Auto-Encoder for reduced dimensions or attributes. The proposed algorithm improved the classification performance significantly. The proposed model was evaluated for efficacy using European credit card data set.
Chapter Objectives
Objectives of this chapter are enlisted below.
- 1.
A brief overview of the credit card.
- 2.
Types of credit card fraud and the losses due to credit card frauds,
- 3.
Examine the many approaches that have been utilized to address the problem of automated fraud detection,
- 4.
introduce a two-step fraud detection system’s innovative architecture and components,
- 5.
A description of the experiment’s design, evaluation criteria, and results,
- 6.
Examine the trends and challenges in identifying credit card fraud.
Key Terms in this Chapter
Autoencoder: Autoencoder is a kind of artificial neural network. It is used to learn unsupervised efficient data encoding.
Batch Normalization: Batch normalization is a deep neural network training strategy that quantifies the inputs per layer of each mini-batch. It serves to maintain the learning process in place. It also reduces the number of epochs required to train deeper networks.
Decision Trees: It is a supervised learning model. It learns the decision rules from the training data to classify an unknown test example.
Clustering: Clustering is an unsupervised learning technique wherein the intragroup similarity is the maximum, and inter-group similarity is the minimum.
Artificial Neural Network: Artificial neural networks (ANNs) are computer systems that imitate an animal’s biological neural networks.
Classification: A data mining task that builds a predictive model to predict the target label of an unknown test record.
Curse of Dimensionality: Difficulty in analyzing the dataset becomes increasingly complicated as the number of dimensions rises. It is known as the curse of dimensionality.
Dimensionality: The number of qualities considered for evaluation while studying an object is called dimensionality.
Credit Card: A credit card is a financial tool that enables cardholders to carry out a credit-based transaction.