Survey On Credit Card Fraud Detection Information Technology Essay

1920 words (8 pages) Essay

1st Jan 1970 Information Technology Reference this


Disclaimer: This work has been submitted by a university student. This is not an example of the work produced by our Essay Writing Service. You can view samples of our professional work here.

Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of

The advent of credit card increases the people comfort but also attracts fraudsters. Credit cards are good targets for fraud, because in a short time large amount of money can be earned without taking risks. The crime will be discovered after few weeks so it is easy for malicious agents to commit this crime. For the past 20 years financial organizations have seen increase in the amount and types of fraud.

The best method is to testify the reasons of fraud from the available data. From several researches the solutions for this credit card fraud are determined by genetic algorithms, artificial intelligence, artificial immune systems, visualization, database, distributed and parallel computing, fuzzy logic, neural networks and pattern recognition. There are many specialized fraud detection solutions which protects credit card, insurance, retail, telecommunications industries. The main objective of these detection systems is to identify the trends of fraudulent transactions.

In this paper Section2 describes Problems in Credit card Fraud detection. Section 3 describes Credit Card Fraud detection using Web services based scheme. Section4 describes Credit Card Fraud detection using neural networks. Section5 describes Hidden Markov Model in Credit Card Fraud Detection. Section6 describes BLAST-SSAHA Hybridization.

In knowledge discovery several techniques like case based reasoning, decision tree, and neural network are used to develop fraud detection systems. These techniques should know about the normal and fraud transactions to identify fraud patterns. For a bank the ratio of fraud transactions is less compared to normal transactions. The fraudsters often initiate new fraud attacks. If a bank cannot regularly update its fraud patterns, it may suffer from fraud attacks. For this problem data-sharing is the best solution. If banks can share their individual fraud transactions to a central center, the integrated data will extract more fraud patterns to help banks improving their fraud detection. But the distributed fraud transactions in banks could be stored in various formats. Because of the formats of fraud transactions the banks cannot exchange data.

To solve this problem, web services techniques, such as XML, SOAP, and WSDL , is established to smooth channel of data exchange across heterogeneous applications of banks. In this data mining techniques are also included to develop a collaborative fraud patterns mining service for credit card fraud detection. To establish data exchange all participant banks should follow the uniform data formats. For exchanging data FPMSC publishes a WSDL file. The WSDL file describes the implementation and interface specification of its provided service. Banks must obey the regulations in the WSDL file. The input messages are fraud transactions and output messages are fraud patterns. Transaction schema is specific schema for input and it is based on XML Schema . Patterns-schema is specific schema for output and it is based on PMML. PMML is a markup language based on DTD standard for defining the predictive models produced by data mining systems.

When participant banks want to access the collaborative fraud patterns they transform their fraud transactions in XML. The XML document passes through the validation by Patterns-schema. The valid xml document is enveloped in SOAP. The SOAP Message can be sent to FPMSC via popular protocols such as HTTP, SMTP, and MIME. The SOAP Message sent to FPMSC can be accumulated into the integrated fraud transactions.

FPMSC extracts fraud patterns using Fraud Patterns Mining (FPM) algorithm. FPM algorithm is developed based on Apriori algorithm. FPM algorithm is used to find fraud patterns. The fraud patterns are transformed to a PMML document validated by Patterns-schema. The PMML document is then enveloped within SOAP Message. The participant banks receive the SOAP message and interpret the PMML document. The participant banks use these fraud patterns, to avoid attacks


3.1 Advantages and Disadvantages of Web services based scheme:

The advantage of this technique is the patterns are represented in rules. So other banks cannot decode these patterns. The data of the participant bank cannot exposed to other banks. This technique applies in both web services and data mining techniques.But this technique cannot used to detect new frauds. Using this technique we can establish a collaborative scheme for credit card fraud detection. In a distributed and heterogeneous environment the participant bank can share fraud patterns. Future enhancement in this technique is web services-based collaborative scheme can be modified as a knowledge sharing mechanism to different enterprises and industries[1].

Fig1 The architecture of the web services based collaborative scheme.

4 Credit Card Fraud detection using neural networks

This system is based on neural network credit card fraud detection and provides interface for commercial databases. The main objective of this technique is to train a neural network with the past data of a particular customer. Then permit the network process the current spending patterns to detect possible anomalies. The thief want to buy things as much as possible. On that time some credit card like VISA, AT&T uses neural network and raise alarm bells when an unusual spending pattern occurs on a customer’s credit card account.

4.1 Advantages

The advantage of this technique is it can able to work on commercial databases without capacity limitations. It has a sophisticated graphical user interface. It is easily extensible to integrate with other intelligent techniques for credit card fraud detection. The performance is satisfactory.

The Existing System are Clementine by Integral Solutions Ltd. features decision tree induction and neural networks, the K-Nearest Neighbor (KNN) method, genetic algorithms, and parallel technology, Intelligent Miner by IBM Corporation supports clustering with demographic and neural clustering . The success rates and performance of all these existing techniques are less. In this technique the purchasing pattern is generated by a tool called transaction generator. The input to the transaction generator is the category of a purchase, information about the “typical” amount of money spent for this category, and information about the “normal” time passed since the last purchase of the same category has taken place. By using these input the fraud was detected. The future enhancement is we can adapt the system to parallel databases, and we can improve GUI to make the control of the system more intuitive. Cardwatch can be extend to general purpose anomaly detection. The limitation in Cardwatch is “one network per customer” should be eliminated. We can introduce more comfortable neural network techniques in Cardwatch[2].

5 Hidden Markov Model in Credit Card Fraud Detection

HMM model is trained by normal behavior of card holder. The sequence of operations is sent to HMM. If the HMM is not accepting the transaction means it is called fraudulent transaction. HMM wont reject the actual transaction. Analysing the behavior of existing purchase data of cardholder fraud is detected.

5.2 Problems in Existing System

The existing system like Cardwatch, web services based fraud detection needs labeled data for both genuine and fraudulent transactions. New frauds cannot be found in these existing techniques.

5.3 Advantages of HMM:

It does not requires fraud signatures.

It detects frauds by considering cardholders purchasing data. The items purchased in individual transactions are not known to a fraud detection system running at the bank that issues credit cards to the cardholders. The HMM-based approach is reduction in the number of false positives – transactions identified as malicious by an FDS although they are actually genuine.No other published literature on the application of HMM for credit card fraud detection. The system is scalable for handling large volumes of transactions[3].

6 Dempster-Shafer theory and Bayesian learning

This technique combines different types of information from current and past behavior. The fraud detection system (FDS) consists of components such as filter and adder.The incoming transaction’s suspicion level is determined by rule-based filter. Multiple evidences is combined using. By finding the initial belief abnormal transactions are classified.

By finding the similarity between genuine and fraud transcations the fraud is detected using Bayesian.

6.1 Advantages of Dempster-Shafer theory:

Using Extensive simulation with stochastic models the performance is increased. If we compare F. Cremer, Sensor data fusion for anti-personnel landmine detection to Dempster-Shafer theory and Bayesian learning testing on synthetic data set, Dempster-Shafer and Bayes approach outperform the fuzzy technique. Aleskerov et al. [2] tested the performance of their CARDWATCH system on sets of synthetic data based on Gaussian distribution only. Chan et al. [5] have used skewed distribution to generate a training set of labeled transactions. None of the techique combine appropriate distributions for generating both the transaction amount and the time gap. This technique has significant improvement in accuracy. Dempster-Shafer theory gives good performance, in terms of true positives and Bayesian learning helps to further improve the system accuracy[4]. The system can be further enhanced by combining conflicting evidences. We could cluster the transaction gaps to determine separate Di’s for each cardholder[4].

7 Fraud detection using anomaly detection and misuse detection

In this technique both anomaly detection and misuse detection techniques are combined. For sequence alignment two analyzer are there namely profile analyzer (PA) and deviation analyzer (DA). The resemblance between incoming and past sequence is analyzed by PA. The profile analyzer traces the unusual transactions. This is sent to DA for alignment. By comparing with past fraudulent transaction the alignment is done. From the two analyzers result the nature of the transaction is determined.

To achieve on-line response time for both PA and DA, the two sequence alignment algorithms BLAST and SSAHA are combined. Hybridization of BLAST and SSAHA results in increased speed of processing. In the proposed method a sequence alignment based credit card fraud detection system which is referred as BLASTFDS[6]. BLAST determines the resemblance between incoming and past sequence. Temporal spending behavior is not captured in BLAST. BLAST takes substantially long time in the alignment process.

The alignment process needs to be fast because millions of cardholders under a particular issuing bank to achieve a satisfactory level of customer service. so we introduce a new algorithm BLAH that combines the advantages of BLAST and SSAHA algorithms. The BLAH algorithm is used to develop a fraud detection system named BLAHFDS. Sequences containing time and amount information are merged together to form a sequence in time-amount dimension. This technique gives significant improvement in accuracy over the systems that consider each dimension individually [7].

Cite This Work

To export a reference to this article please select a referencing stye below:

Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.

Related Services

View all

DMCA / Removal Request

If you are the original writer of this essay and no longer wish to have your work published on the website then please:

Related Lectures

Study for free with our range of university lectures!