Credit Card Fraud Detection Using Machine Learning

Unprecedented advancement of e-commerce soars the frequency of online and offline financial transactions of Credit Card as a popular means of payment for public. With the tremendous frequency of transactions per minute worldwide, the multi-fold risk of fraudulent transaction has increased significantly for both the parties either user or issuer. This paper presents the comprehensive survey on multiple machine learning approaches to credit card fraud detection (CCFD). The existing approaches are eliciting good responses in terms of accuracy but the precocious Deep Learning algorithm (here, Convolutional Neural Network) was deployed in the anticipation of better accuracy. In this paper, comparative analysis has been carried out among various Machine Learning algorithms. Analytical parameters such as counts of layers, epochs & models have been employed. Outlandish outcome found for various machine learning classifier algorithms such as Random Forest, Support Vector Machine, K-Nearest Neighbor, Gaussian Naïve Bayes, Decision Tree, Logistic Regression, moreover, the dataset was fed to Convolutional Neural Network (CNN). The performance metrics for aforesaid classifiers in accordance with standard criteria was recorded. The best outcome was found with Random Forest Classifier depicting F1-score as 85.71%, Precision as 97.40%, and Accuracy as 99.96%.


INTRODUCTION
Credit card fraud indicates to the unauthorized use of someone else's credit card information to make purchases or carry out fraudulent transactions.This type of fraud can occur through various means, and perpetrators often aim to exploit vulnerabilities in the credit card system for financial gain.Transaction involves several stages as shown in figure 1, from the initiation of a purchase to the authorization and settlement of the transaction.Customer Initiates a Purchase then the process begins when a customer decides to make a purchase using a credit card.This can occur at a physical point of sale (POS) terminal or online through an e-commerce website [1]- [4].As a next step there is merchant submission, for in-person transactions, the merchant (business or service provider) swipes, inserts, or taps the customer's credit card using a card reader at the POS terminal.For online transactions, user enters their CC details on the website's payment page.It leads to authorization request, the merchant's payment system sends an authorization request to the credit card issuer (bank or financial institution that issued the credit card) to confirm whether the user has a balance of sufficient credits and the transaction request is valid.It needs an authorization approval; the credit card issuer reviews the authorization request, checks the customer's credit limit, and assesses the transaction's validity.If approved, the issuer sends an authorization code back to the merchant.It moves to transaction approval at merchant; upon receiving the authorization code, the merchant's system approves the transaction, and the customer is informed that the purchase has been successful.A record of the transaction details, including the authorization code, is stored by both the merchant and the credit card issuer for future reference.Throughout the day, the merchant accumulates authorized transactions and submits them in a batch to the acquiring bank or payment processor [5]- [8].This is known as batch processing.The acquiring bank or payment processor forwards the batch of authorized transactions to the credit card network (such as Visa or MasterCard) for clearing.Clearing involves the exchange of transaction data between the acquiring and issuing banks.The issuing bank receives the details of the authorized transactions from the credit card network.If the transaction is approved, the issuing bank transfers funds to the acquiring bank. r

Figure 1: Credit card process of financial transaction
This assures that the merchant gets paid for the services or goods provided to the customer.It is to emphasize that the credit card transaction process involves multiple parties, including the cardholder, merchant, acquiring bank, credit card network, and issuing bank.Additionally, security measures such as encryption and tokenization are implemented to protect sensitive cardholder information during the transaction process.Here, in process there lies a high vulnerability of credit card frauds.
Figure 2 shows few key areas of credit card fraud.

Authorization Response
Customer provides credit card information to the merchant to make a purchase.
Merchant submits transaction details to its acquiring bank (merchant bank) for authorization.
Acquiring bank forwards authorization request to the credit card issuer.
Credit card issuer evaluates the request based on credit limit, history, and fraud indicators, sending an authorization code if Acquiring bank receives the authorization response and communicates it to the merchant.

Merchant Receives Payment
Merchant captures transaction details, and the acquiring bank initiates the process of settling funds.
Acquiring bank sends transaction details to the credit card network (Visa, Mastercard, etc.) for clearing.

Credit card network forwards transaction details
to the issuing bank for settlement.
Issuing bank transfers funds to the acquiring bank for settlement.
Acquiring bank deposits funds into the merchant's account, completing the credit card payment process.
Criminals use skimming devices to capture data from the magnetic stripe on credit cards.These are often implanted on ATMs, gas pumps, or point-of-sale terminals.
Fraudsters broadcast deceitful emails, messages, or websites that appear legitimate, tricking individuals into provisioning their credit card information.Cybercriminals gain unauthorized accessibility to a user's account, often through phishing or hacking, and use the victim's credit card for fraudulent transactions.Criminals may use lost or stolen credit cards to make illegitimate purchases before the cardholder can report the loss.Criminals use pilfered credit card details to make small online purchases to test whether the card is still active before making larger transactions.Fraudsters file applications for credit cards using stolen or fabricated personal information to open new accounts.Similar to card skimming, criminals use software to generate fake credit card numbers, which may be used for online transactions.
Table 1 shows the various aspects associated with the credit card fraud and evaluating impacts on all three parties card holder, issuing bank and point of sale.These become very severe and vulnerable over the level of finance associated with the fraud.Cybercriminals gain unauthorized access to a user's account, often through phishing or hacking, and use the victim's credit card for fraudulent transactions.Lost or Criminals may use lost or stolen credit cards to make unauthorized purchases before the cardholder can report the loss.Criminals use stolen credit card information to make small online purchases to test whether the card is still active before making larger transactions.Fraudsters place the applications for credit cards using stolen or fabricated personal information to open new accounts.Similar to card skimming, criminals may install skimming devices on point-of-sale terminals to capture credit card data during transactions.Criminals use software to generate fake credit card numbers, which may be utilized for online transactions.Fraudsters manipulate individuals into revealing their credit card information through psychological tactics or impersonation.Fraudsters create a fake online store and use stolen credit card data to purchase goods from a legitimate store, having the goods shipped to the victim's address [15]- [18].Legitimate cardholders dispute valid transactions with their credit card issuer, often claiming the transaction was unauthorized, resulting in chargebacks.Criminals steal credit card statements or new credit cards from the victim's mailbox.Fraudsters create a fabricated scenario to receive personal data, including credit card details, from the victim.Cybercriminals intercept and alter interaction between the user and a legitimate website, capturing credit card info.Malicious software installed on a user's device can capture such details inputted during online transactions.Figure 3 shows relative analysis of various frauds occurred in credit card financial systems.In other words, Credit Card Fraud may be described as a case where a person uses someone else's credit card for personal reasons while the owner and the card issuing authorities are unaware of the fact that the card is being used.A stolen, lost or fake Credit Card could lead to fraud [20]- [24].Dealing with imbalanced datasets where the count of legitimate transactions far exceeds fraudulent ones, affecting the criterion of machine learning models.

Emerging Fraud Patterns
Staying ahead of fraudsters who continually innovate and adapt their tactics, challenging traditional rule-based systems to keep up with evolving fraud patterns.Cross-Channel Fraud Detecting fraud across diverse channels such as online, mobile, and inperson transactions, and integrating data from different sources.

Synthetic Identities
Identifying and preventing fraud involving synthetic identities created by combining real and fake information in credit card applications.

Insider Threats
Mitigating the risk of collusion or insider involvement in fraudulent activities, as insiders may have access to sensitive information.

Global Transactions
Monitoring and identifying fraudulent activities across borders, dealing with variations in regulations, transaction patterns, and data sources in different regions.

Data Privacy Concerns
Balancing effective fraud detection with user privacy considerations, implementing robust security measures, and complying with data protection regulations.

Dynamic Fraud Schemes
Adapting to fraudsters' quick changes in tactics and keeping in sync with technological advancements, requiring constant innovation in fraud detection methods and technologies.Technological Challenges Overcoming challenges associated with implementing and integrating new technologies, especially for organizations with legacy systems.As world is going toward cashless economy, usage of Credit or Debit Card in e-shopping is increasing day by day, and so associated frauds causing huge monetary loss.Credit card fraud can be divided into two main types: application fraud and behavioural fraud.Both types involve various tactics and techniques used by fraudsters to exploit vulnerabilities in the credit card system.Table 2 shows major challenges associated with the credit card fraud detection [25]- [30].

A. Aspplication Fraud:
Identity Theft: Fraudsters may use stolen or fabricated personal information to file applcation for credit cards in someone else's name.Synthetic Identity Fraud; criminals create fictional identities by combining real and fake information to apply for credit.Over time, they build up the creditworthiness of these synthetic identities before exploiting them.Collusion; this involves an individual or group working with an insider, such as a corrupt bank employee, to submit fraudulent credit card applications.Document forgery: Criminals may forge documents, such as pay stubs or utility bills, to support their credit card applications.

B. Behavioural Fraud:
Stolen Card Fraud: Criminals gain access to credit card information and use it for unauthorized transactions.Account Takeover; fraudsters gain control of a legitimate cardholder's account through various means, such as phishing or hacking, and make unauthorized transactions.Card Not Present (CNP) Fraud; this occurs when the physical card is not required for a transaction, such as online or overthe-phone purchases.Fraudsters may use stolen card details for these transactions.Skimming; criminals use devices called skimmers to collect credit card information from the magnetic stripe when a card is swiped at an ATM or point-of-sale terminal.Phishing and Social Engineering; fraudsters use deceptive emails, messages, or phone calls to trick individuals into providing their credit card information.

C. Preventive Measures:
Verification and Authentication: Implement robust identity verification processes during the application phase.This can include document verification, biometric authentication, and other advanced identity verification methods.Machine Learning and AI: Employ advanced technologies to analyse patterns of behaviour and to detect anomalies that may point out fraudulent activity.Encryption and Tokenization: Protect cardholder data by encrypting sensitive information and replacing it with tokens, making it harder for fraudsters to gain access to valuable data.Multi-Factor Authentication: Implement multifactor authentication measures for online transactions to add an extra layer of security.Credit card issuers, financial institutions, and consumers all play a role in preventing and mitigating credit card fraud by staying vigilant and adopting best practices in security and fraud detection.

CONVENTIONAL METHODS OF CREDIT CARD FRAUD DETECTION
Conventional methods of credit card fraud detection typically involve rule-based systems, heuristics, and predefined patterns to identify potentially fraudulent transactions.While these methods have been effective to some extent, they may struggle to keep pace with the evolving tactics of fraudsters.Figure 2 shows factors affecting the process effectiveness of CCFD.[31]- [36]. A. Rule-Based Systems: Description: Rule-based systems employ predefined rules and conditions to flag transactions that match specific patterns associated with known fraud.Pros: Simple to implement, easy to understand, and can quickly identify known fraud patterns.While these conventional methods provide a baseline for credit card fraud detection, the industry is increasingly turning to advanced technologies, such as machine learning, artificial intelligence, and behavioural analytics, to enhance the efficiency and accuracy of fraud detection systems.These technologies can get adapted to evolving fraud patterns and deliver more accurate and timely identification of suspicious activities as depicted in figure 4. Further an important aspect is the considering main features to determine the fraud and its frequency.The threat may appear through any feature from the process of CC transaction as deliberated in table 3 [36]- [38].

PROPOSED METHODOLOGY
Hybrid approaches combine supervised and unsupervised learning techniques to enhance fraud detection accuracy.By leveraging the strengths of both methods, these methodologies can generate more robust and accurate results.One example is the semi-supervised learning approach, where a small portion of labeled fraud data is used to train a supervised learning model.This model is then utilized to detect fraud in the unlabeled data.By combining the power of labeled data with the ability to identify anomalies in unlabeled data, hybrid approaches can improve the overall effectiveness of fraud detection systems.Deep learning techniques, especially neural networks have gained significant popularity in recent years for credit card fraud detection.These models can automatically learn complex patterns and relationships in the data, making them extremely effective in detecting fraudulent transactions.Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) have shown promising results in fraud detection tasks.By leveraging the hierarchical structure of neural networks, these models can capture intricate patterns and identify subtle indicators of fraud.The performance of credit card fraud detection models is typically measured using evaluation metrics such as accuracy, precision, recall, and F1-score.These metrics offer insights into the model's ability to accurately classify transactions as fraudulent or legitimate.Additionally, techniques like cross-validation and Receiver Operating Characteristic (ROC) curves help assess the model's generalization and robustness.By evaluating the models using these metrics, businesses and financial institutions can have informed decisions about the effectiveness of their fraud detection systems [41]- [42].

Customer Authentication Solutions
Deploying advanced customer authentication methods (e.g., twofactor authentication, tokenization) to enhance transaction security.

Continuous Learning Systems
Implementing systems that continuously learn and adapt to new fraud patterns, reducing the need for frequent rule updates and staying ahead of emerging threats.

Table 6: Enhanced machine learning methods used for CCFD Reference
Methodology and Key Points [8] Unsupervised feature learning using a stacked sparse autoencoder (SSAE) for fraud prediction.[9] Neural network ensemble classifier with a hybrid data resampling method.
Base learner: Long short-term memory (LSTM) in adaptive boosting (AdaBoost) technique.[7] Credit card fraud detection using ML algorithms (RF, NB, MLP) and Synthetic Minority Oversampling Technique (SMOTE) for imbalanced data.RF algorithm showed the highest accuracy.[10] Feature selection to minimize data overlap using algorithms (RONS, ROS, ROA) built through sparse feature selection.Binary classification with good performance.[11] Combination of oversampling and feature selection methods to improve classification algorithms.Significant performance improvement demonstrated.[7] Intelligent payment card fraud detection system.Assessment of aggregated features identified by a genetic algorithm for improved fraud detection accuracy.
[12] Hybrid approach using Recursive Features Elimination (RFE), Hyper-Parameters Optimization (HPO), and SMOTE.Excellent performance across different datasets.[13] Data-point machine learning with SMOTE-based oversampling.Various classifiers tested, showing increased accuracy for identifying fraudulent transactions.[14] Feature selection using Enhanced Neural Networks (ENN) and Artificial Bee Colonies (ABCs) for improved accuracy in credit card fraud classifications.Logical relationships explored through LGBPs.[6] Two-stage approach: selection of optimal ML algorithms (LR, KNN, DT, NB, RF, GBM, Light GBM, XG Boost, Cat Boost) and integration with different resampling techniques.AllKNN-Cat Boost outperformed.[15] SVM hyper parameter optimization (c and sigma) using Cuckoo Search Algorithm, Genetic Algorithms, and Particle Swarm Optimization.Recommendation for exploring new algorithms in future work.

CONCLUSION
Machine learning techniques have revolutionized credit card fraud detection by enabling businesses and financial institutions to detect and prevent fraudulent transactions in real-time.This has provided a comprehensive review of various machine learning approaches, including supervised learning, unsupervised learning, hybrid approaches, and deep learning techniques.It is crucial to consider the strengths and limitations of each technique and choose the most appropriate approach based on the available data and resources.With ongoing advancements in machine learning, credit card fraud detection will continue to evolve, making it increasingly challenging for fraudsters to exploit the system.By staying up-to-date with the latest techniques and continuously improving fraud detection systems, businesses and financial institutions can effectively combat credit card fraud and protect their customers' financial security.Credit Card fraud is without a doubt an act of criminal dishonesty.It is a financial threat to both credit card issuing companies & its holders.Out of millions of transactions in a fraction of time, a robust classifier is required to distinguish between fraud transactions and non-fraud ones.
Overall, credit card fraud detection is an ongoing challenge that requires continuous research and innovation.By leveraging the power of machine learning and privacy-preserving techniques, we can enhance the security of financial transactions and protect customers from fraudulent activities.

Figure 2 :
Figure 2: Major domains of credit card fraud

Figure 3 :
Figure 3: Relative analysis of various types of fraud detection Cons: Limited adaptability to new and emerging fraud schemes; may generate false positives or negatives.B. Transaction Monitoring: Description: Transaction-monitoring for uncommon patterns or deviations from a cardholder's normal spending behavior.Pros: Can identify anomalies based on historical transaction data.Cons: May trigger false alarms for legitimate but uncommon transactions; may not detect sophisticated fraud patterns.C. Address Verification System (AVS): Description: Verifying the billing address provided during a transaction against the one on file with the card issuer.Pros: Adds an additional layer of verification.Cons: Limited effectiveness in cases where billing information has been compromised; doesn't address online transactions without address verification.D. Card Verification Code (CVC) Check: Description: Verifying the three-digit code on the back of the credit card during transactions.Pros: Adds an extra layer of security for online and card-not-present transactions.Cons: Doesn't prevent fraud if the card information has been compromised; limited effectiveness in some scenarios.E. Velocity Checks: Description: Monitoring the frequency and volume of transactions within a specified timeframe.Pros: Can identify unusual patterns such as a sudden increase in transaction frequency.Cons: May initiate false positives for legitimate high-frequency transactions.F. Geolocation Checks: Description: Verifying the location of the transaction against the cardholder's usual locations.Pros: Adds a layer of verification based on geographical information.Cons: Limited effectiveness for individuals who frequently travel; may not detect remote or online transactions.G. Manual Review and Investigation: Description: Human review of flagged transactions for further investigation based on suspicion or predefined criteria.Pros: Allows for nuanced judgment and investigation of complex cases.Cons: Resource-intensive and time-consuming; may delay transaction approvals.

Figure 4 :
Figure 4: Factors affecting the process effectiveness of credit card fraud detection b) Feature engineering: Extract relevant features from the dataset that can help in recognizing fraudulent and legitimate transactions.This may involve creating new attributes based on domain knowledge or using techniques such as dimensionality reduction.c) Model selection: Choose an appropriate machine learning algorithm for fraud detection.Commonly used algorithms include logistic regression, decision trees, random forests, and neural networks.Consider the trade-offs between accuracy, interpretability, and computational complexity when selecting the model.Model training: Split the preprocessed dataset into training and testing sets.Train the selected machine learning model on the training set, using the features and the corresponding labels (fraud or non-fraud).Model evaluation: Evaluate the trained model on the testing set to measure its performance.Common evaluation metrics include accuracy, precision, recall, and F1 score.Adjust the model parameters or try different algorithms if the performance is not satisfactory.d) Model deployment: If the model performance is satisfactory, deploy it into production.This involves integrating the model into the credit card payment system and continuously monitoring incoming transactions for potential fraud.e) Model maintenance: Periodically retrain and update the model using new data to adapt to changing fraud patterns.Monitor the performance of model and make necessary adjustments to ensure its effectiveness.It's important to note that the above steps provide a general framework for credit card fraud detection using machine learning.The specific implementation details may vary depending on the dataset, algorithms, and tools used.

Figure 5 :
Figure 5: Process flow of credit card fraud detection using machine learning

Figure 6 :
Figure 6: Relative analysis of various types of fraud detections

Figure 7 :
Figure 7: Advanced soft computing process flow

Figure 8 :
Figure 8: ML and DL approaches for CCFD

Figure 9 :
Figure 9: Various advanced of CCFD techniques

Table 1 : Descriptive investigation of credit card fraud ASPECTS DESCRIPTION Methods
of Fraud Skimming, phishing, lost or stolen cards, data breaches, carding, identity theft, account takeover, etc. Stolen Information Criminals use skimmers, phishing emails, or deceptive methods to obtain credit card details.

Table 3 : Feature affecting credit card fraud detection S.No. Name of Feature
Supervised ML algorithm like SVM is employed for solving linear and non-linear classification problem of dataset like Image Recognition, Credit Rating etc. SVM creates Hyper plane to separate input data in support vector.Although, several Deep Learning algorithms are available namely having application in computer vision, NLP, Heart disease detection, healthcare fraud detection, malware detection, intrusion detection, video surveillance detection, location tracking..In this paper, we are using CNN (Convolutional Neural Networks) for identity theft pertaining to CCFD, whether the transaction is normal or fraudulent one.The process of CCFD as follows: Feature Selection algorithms are rendered to dataset to order the principal-features helping to make predictions based on classification.Feature-Extraction algorithms, using Deep Learning model, are deployed to extract/generate new features and solve clustering problem from the dataset of CCFD.The performance of CNN model is analyzed by adding layers.The comparative-assessment between ML & DL is carried out.The outcome shows that CNN model proposed outnumbers the ML model.To examine the accuracy of classifier in ML and clustering in DL, the Model Performance Evaluation benchmarks such as F1-scor, Precision, AUC curve and Accuracy [40]NCED METHODS OF CREDIT CARD FRAUD DETECTIONAs per record, there were 393207 instances of such frauds out of 1.4 million identity theft reported ones.Therefore, an Automated CCFD comes as a rescue and important tool for financial institutions.The CCFD using ML model is a supervised (classification) model to recognize fraudulent and non-fraudulent transactions.The ML model deals with various underlying problems such as system-reaction-time, costsensitivity, pre-processing of features (PCA etc.) to predict as per prior data patterns.areemployed.The most recent dataset are used to carry out the experiments.The common procedure behind the process is shown in figure5[39]-[40].a) Data preprocessing: Clean and preprocess the collected data.This involves eliminating duplicates, treating missing values, and normalizing or scaling the numerical features.It is also crucial to balance the dataset to ensure equal representation of both fraud and non-fraud cases.