Online and credit card payments play a significant role in today’s economy. They have become an unavoidable part of household, business, and global activities.
However, ever since the introduction of digital payments, scammers have found ways to undertake fraudulent transactions and unauthorized purchases, using the credit card information stolen from the users.
These fraudulent activities have become a growing problem in recent years for individuals using credit cards and merchants, banks, eCommerce websites, and other financial organizations.
Credit card fraud is a form of financial fraud involving fake or stolen credit card information and causing financial harm to account holders or merchants. It is increasing significantly with the development of modern technology, resulting in billions of dollars each year.
Credit card fraud tricks belong mainly to two groups: application and behavioral fraud. Application fraud occurs when fraudsters apply for new cards from banks or issuing companies, using false or other’s information (called identity fraud). Behavioral fraud has four principal types: stolen/lost card, mail theft, counterfeit card, and ‘cardholder not present’ fraud.
An efficient fraud detection involves identifying scarce fraud activities among numerous legitimate transactions as quickly as possible. However, fraud detection is essentially a rare event problem, which has been variously called outlier analysis, anomaly detection, exception mining, imbalanced mining data, etc.
Since the number of fraudulent transactions is usually a meager fraction of the total transactions, the task of detecting fraud transactions accurately and efficiently is relatively tricky and challenging. Many techniques are used to confront the growing payment fraud, but each has its own drawbacks, advantages, and characteristics.
Challenges in payment fraud detection
- Data imbalance: The credit card fraud detection data usually has imbalanced nature. It means that minimal percentages of all credit card transactions are fraudulent. This makes the detection of fraud transactions complicated and imprecise.
- Different misclassification importance: In a fraud detection task, various misclassification errors have varying significance. Misclassifying a normal transaction as fraud is not as harmful as detecting a fraud transaction as usual.
- Overlapping data: Many transactions may be considered fraudulent, while actually, they are normal (false positive), and reversely, a fraudulent transaction may also seem legitimate (false negative). Hence obtaining a low rate of a false positive and false negative is a key challenge.
- Lack of adaptability: When it comes to fraud detection, the classification algorithms are usually faced with detecting new types of normal or fraudulent patterns. The supervised and unsupervised fraud detection systems are often inefficient in detecting new patterns of normal and fraud behaviors.
- Non-existence of standard: Credit cards are inherently private property, so creating a proper benchmark for this purpose is very difficult.
- Lack of suitable metrics: The limitation of good metrics to evaluate the fraud detection system results is an open issue.
Machine learning is considered one of the most successful techniques to identify fraudulent transactions and prevent them from happening while using banking services like credit cards, Automated Teller Machines (ATM), internet and mobile banking services.
Benefits of machine learning in fraud detection
- Scalable and highly accurate results
- Fewer false positives, through sophisticated behavioral analysis
- Reduce the need for manual review
- Ability to stop fraud without impeding the user experience
- Proactive, i.e., it can tell you what’s happening
- Frees up the tech teams’ time to focus on more strategic tasks
- Lower operational costs than other approaches
- Can be automated
- Adapts quickly
- Learn across many data elements
All credit card fraud detection techniques are classified into two general categories: fraud analysis (misuse detection) and user behavior analysis (anomaly detection). The first group of techniques deals with supervised classification tasks at the transaction level, while the second approach deals with unsupervised methodologies based on account behavior.
Let’s now discuss different fraud detection techniques using machine learning and compare them using performance measures like accuracy, precision, and specificity.
1. Artificial Neural Network (ANN)
An artificial neural network (ANN) is a set of interconnected nodes designed to imitate the functioning of the human brain. ANNs can be configured by supervised, unsupervised, or hybrid learning methods. In supervised learning, samples of fraudulent and non-fraudulent records associated with their labels are used to create models. These techniques are often used in the fraud analysis approach. The unsupervised techniques do not need the previous knowledge of fraudulent and normal records. These methods raise the alarm for those transactions that are most dissimilar from the normal ones. One of the advantages of using unsupervised neural networks is that they can learn from the data stream.
- Ability to learn from the past
- Ability to extract rules and predict future activities based on the current situation
- Lack of need to be reprogrammed
- High accuracy
- High speed in detection
- The ability to generate code to be used in real-time systems
- The easiness to be built and operated
- Effectiveness in dealing with noisy data, in predicting patterns, in solving complex problems, and in processing new instances
- Adaptability and maintainability
- Difficulty in confirming the structure
- Poor explanation capability
- High processing time for large neural networks and excessive training
- Difficult to set up and operate
- High expense
- Non-numerical data need to be converted and normalized
- Sensitivity to data format.
2. Artificial Immune System (AIS)
Artificial Immune System (AIS) is a recent subfield based on the biological metaphor of the immune system, distinguishing between self and non-self-cells, or more specifically, between harmful cells (called pathogens) and other cells. Researchers use immunology concepts to develop a set of algorithms, such as negative selection algorithm, immune networks algorithm, clonal selection algorithm, and the dendritic cells algorithm.
- High capability in pattern recognition
- Powerful in learning and memory
- Easy in integration with other systems
- Dynamically changing coverage
- Has diversity
- Noise tolerance
- Fault tolerance
- Predator-prey dynamics
- No need for a training phase in DCA.
- Poor in handle missing data in ClonalG and NSA
- Need high training time in NSA
3. Genetic Algorithm
Genetic algorithms have been used in data mining tasks, mainly for feature selection. It is also widely used in combination with other algorithms for parameter tuning and optimization. The availability of genetic algorithm code in different programming languages is a popular and strong algorithm in credit card fraud detection.
- Works well with noisy data
- Usually combined into other techniques to increase the performance of those techniques and optimize their parameters
- Easy to integrate with other systems
- Easy in build and operate
- Fast in detection
- Adaptability and maintainability
- Knowledge discovery and data mining
- Requires extensive tool knowledge to set up and operate and difficult to understand.
- Very expensive in consuming time and memory.
4. Hidden Markov Model (HMM)
A Hidden Markov Model is a double embedded stochastic process applied to model much more complicated stochastic processes than a traditional Markov model. The underlying system is assumed to be a Markov process with unobserved states. In credit card fraud detection, an HMM is trained for modeling the normal behavior encoded in user profiles. According to this model, a new incoming transaction will be classified as fraud if it is not accepted by the model with a sufficiently high probability. Each user profile contains a set of information about the last 10 transactions of that user lifetime, category, and the amount for each transaction.
- Fast in detection
- Detects credit card frauds with a low false alarm.
- Scalable for processing a huge number of transactions.
- Highly expensive
- Low accuracy
- Not scalable
- Too large size data sets
5. Support Vector Machines (SVM)
Support vector machine (SVM) is a supervised learning model with associated learning algorithms that can analyze and recognize patterns for classification and regression tasks. SVM is a binary classifier. The basic idea of SVM was to find an optimal hyper-plane that can separate instances of two given classes linearly. SVM is successfully applied to a broad range of applications, including credit card fraud detection.
- Deliver a unique solution since the optimality problem is convex
- Robust, even when the training sample has some bias.
- Poor in process large dataset
- Has a low speed of detection
- Medium accuracy
- Lack of transparency of results
6. Bayesian Network
A Bayesian network is a graphical model that represents conditional dependencies among random variables. The underlying graphical model is in the form of a directed acyclic graph. Bayesian networks are useful for finding unknown probabilities given known probabilities in the presence of uncertainty. Bayesian networks play an important and effective role in modeling situations where some basic information is already known, but incoming data is uncertain or partially unavailable. They are successfully applied to various fields of interest, such as churn prevention in business, pattern recognition in vision, generation of diagnostic in medicine and fault diagnosis, and forecasting in power systems.
- High processing and detection speed
- High accuracy
- Excessive training need
7. Fuzzy Logic Based System
The fuzzy logic-based system is the system based on fuzzy rules. Fuzzy logic systems address the input and output variables’ uncertainty by defining fuzzy sets and numbers to express linguistic variables’ values (e.g., small, medium, and large). Two important types of these systems are fuzzy neural networks and fuzzy Darwinian systems.
- Very fast in detection
- Good accuracy
- Very high accuracy
- Has very low speed in detection
- High expensive
Evaluation criteria for credit card fraud detection
Here, we have some crucial evaluation criteria for credit card fraud detection systems powered by Machine Learning techniques.
- Accuracy (ACC)/Detection rate: Accuracy is the percentage of correctly classified instances. It is the most widely used classification performance metric.
- Precision/Hit rate: Precision is the number of classified positive or fraudulent instances that actually are positive instances.
- True positive rate/Sensitivity: TP (true positive) is the number of correctly classified positive or abnormal instances. TP rate measures how well a classifier can recognize abnormal records. It is also called a sensitivity measure. In the case of credit card fraud detection, abnormal instances are fraudulent transactions.
- True negative rate/Specificity: TN (true negative) is the number of correctly classified negative or normal instances. TN rate measures how well a classifier can recognize normal records. It is also called a specificity measure.
- False-positive rate (FPR): Ratio of credit card fraud detected incorrectly
- ROC: True positive rate plotted against false positive rate. Relative Operating Characteristic curve, a comparison of TPR and FPR as the criterion changes
- F1-measure: Weighted average of the precision and recall