Fraud Detection Techniques

I. INTRODUCTION

With an increase in online shopping, occurrences of credit card fraud are also increasing. Data mining techniques are being called upon to help monitor client accounts for fraudulent activity. There are six steps involved in data mining: one must define the problem, prepare the data, explore the data, build models, explore and validate said models, and then deploy and update the models. Data mining searches established usage patterns to detect unusual pattern types. Unusual pattern types indicate possible fraudulent transactions (Ogwueleka, 2011). 

44 percent of banking transactions now occur online (Hisar, 2014). This makes fraudulent bank transactions easier and more frequent. To combat these transactions, data mining is used. Data mining is the retrieval and observation of data that helps understand legitimate customer usage for purchases versus fraudulent activity.

Data mining consists of two kinds of tasks. Descriptive tasks result in source data being described by patterns. Predictive tasks create models which have predictive abilities (Razak, 2015). Data mining extracts pieces of data from websites or data warehouses. Examination of this data lends understanding to improving data usage (Desai, 2013). 

Data is mined from data warehouses. Data warehouses are multi-dimensional cubes of data (Razak, et al, 2015). Data mining of warehouse data provides a safe environment for past data to be processed and extracted from. This data is used to create algorithms for fraud detection (Redi, 2016).

Emails can also be effective locations for data mining. Data mining can detect spam in emails. These emails are similar in data types to fraud, and there is a large amount of data to draw from in this area (Phua, et al, n.d.).

II. TYPES OF FRAUD

Fraud can be classified into the following categories: card-related fraud, merchant-related fraud, and Internet fraud (Ogwueleka, 2011). When a con artist uses a person’s private information to apply for a credit card in the victim’s name, this is referred to as application fraud. When a credit card is stolen and then used for purchase, that is known as behavioral fraud (Kumari, et al, 2017).

40 percent of financial fraud, as of August 2013, falls under the category of credit card fraud, resulting in a loss of $5.55 billion globally. Committing fraud is becoming even easier by means of applications one can download to a smart phone to skim credit cards (Seeja, et al, 2014). As of 2011, 34 percent of companies across the world reported being victims of fraud, up from 30 percent in 2009 (Akhilomen, 2013).

Credit card usage almost doubled from 2010-2014 and is still on the rise. Besides being a convenient method of payment, credit cards are attractive in that one can purchase something immediately and pay for it later. It is estimated that between 30 and 40 percent of fraudulent credit card transactions can be prevented (Malini, et al, 2017). The average customer does not carry cash anymore, and many places no longer accept checks due to past issues of bounced and fraudulent checks. A fake check can take weeks to be denied, severely affecting multiple banking transactions that occur after the check.

Fraud falls into two basic categories: offline and online. Offline fraud occurs with physical credit card transactions. Often, a bank is able to freeze a card before it can be used. With online fraud, however, it is easier to make multiple transactions before suspicion arises (Desai, et al, 2013).

Credit card fraud does affect the user, but it affects the merchant more (Hisar, et al, 2014). The user is inconvenienced while fighting the charges, waiting for a new card, and waiting for restoration of any missing funds. The merchant, however, must not only spend time, resources, and funds pursuing the fraudster, but must restore the client’s funds at no cost to the client. This financial fallout motivates businesses to deter fraud.

A. Application Fraud

This type of credit card fraud occurs when information is falsified during the credit application process. The fraudster is able to receive a credit card based on someone else’s information. Information concerning individual finances will be falsified under the victim’s name (Pushpalatha, et al, 2017).

B. Assumed Identity

Assumed identity is another common type of fraud. A victim’s information will be used to secure a transaction. The fraudster will use a temporary address to receive stolen goods (Pushpalatha, et al, 2017). This enables the fraudster to evade being caught.

C. Financial Fraud

Financial fraud is different from the methods above. Instead of using someone else’s information, this occurs when financial information about oneself is falsified. The rest of the information will be correct (Pushpalatha, et al, 2017). This increases one’s chances of being approved.

D. Skimming Technology

Skimmers allow fraudsters to copy credit card information (Pushpalatha, et al, 2017). These devices can be placed on Automatic Teller Machines, at gas stations, and in various other locations. Most times, a user does not even realize the skimmer has been placed over the device in use.

E. Never Received

This last type of fraud involves the physical card itself. This happens when the card is en route to the user but is intercepted. The victim never receives the card, and the fraudster activates it and uses it.

III. METHODS OF FRAUD

There are a few different types who benefit from credit card fraud. Credit card information buyers do not have necessary skills to steal information by themselves. These buyers go through websites on the “dark web” to purchase data stolen by others. “Cracker” or “Black Hat” hackers are those who steal credit card information in order to make purchases for themselves (Akhilomen, 2013).

Physical thieves refer to those who steal actual credit cards. These cases are slightly easier to discourage by canceling the card once it has been discovered missing. Credit card fraud generator software and “sniffer” or key logger programs can falsify information, find personal data, or steal authentication information for credit cards, online banking, and website access (Akhilomen, 2013).

Spyware can also be installed on a computer to monitor websites and log ins a user operates. Hackers will sometimes impersonate websites to steal authentication information. The websites can have false purchases set up. Once a user enters credit card information, it is stolen by the hacker for fraudulent use (Akhilomen, 2013).

Computer intrusion is fraud where a computer has been compromised. An outside hacker or a malicious insider may be behind the attacks. There can also be subscription fraud, where a hacker sets up a service using the victim’s information, or super imposed fraud that may show up as charges on a telephone bill (Desai, et al, 2013). 

Intrusion Detection Systems can be used to combat this. It monitors a system and blocks suspicious IP addresses or users. Attacks that can be stopped by Intrusion Detection Systems are probing, denial of service, user to root, and remote to user. Probing occurs when a network is scanned for usable information. Denial of service attacks are when a hacker floods a system with information in an attempt to bring it down. In user to root attacks, a hacker attempts to access information through system roots, and in remote to user attacks, packets are used to exploit a system for access (Redi, 2016).

Transactions made through the Internet or on the phone are known as Card Not Present (CNP) transactions. Using malware, data breaches, phishing, or other methods, a hacker may obtain the card’s data for use in these types of transactions. Since a physical card is unnecessary to make a purchase, this is the preferred method of credit card fraud (Fahmi, et al, 2016).

IV. FRAUD DETECTION METHODS

There are two ways of dealing with credit card fraud: corrective measures and preventative measures. Corrective measures seek to fix fraud that has occurred. Preventative measures are intended to stop the fraud in the first place (Baesens, et al, 2015). Different applications utilize one or both of these measures in various ways.

Data-driven solutions are increasing in popularity. These methods are precise and efficient. Data mining provides cost-effective solutions (Baesens, et al, 2015). The development of applications using data mining techniques help identify fraudulent transactions.

Neural networks are artificial intelligence applications that are a result of data mining. These networks are set up to mimic how the human brain operates and learns. By being trained, neural networks are able to sense a change in consumer patterns to detect fraud (Gayathri, et al, 2013). Neural network applications and algorithms are a leading application in detecting fraudulent transactions (Albashrawi, 2016).

Meta-learning algorithms (algorithms that learn how to learn) interpret the data and learn to identify suspicious activity. Processes that use meta-learning are able to change their behavior based on information acquired as they work. Meta-learning has resulted in the development of a few different applications (Razak, et al, 2015). One way this type of learning can be put to use is in monitoring transactions. A group of clients with low transaction amounts trigger suspicion if a large withdrawal is made, setting off the neural network’s alarm system and notifying banking agents (Soltaniziba, et al, 2015).

A. KNN

K-Nearest Neighbor is based on instance-based learning. Instances are stored in this application and judged against new instances. A distance metric is used (nearest neighbor) to determine if transactions near another fraudulent transaction may also be fraudulent. Comparing transactions shows any abnormal patterns in customer behavior. This application exhibits consistently high performance (Seeja, et al, 2014).

B. Naïve Bayes

Naïve Bayes assumes that attributes are independent, resulting in its name. To this method, the presence or absence of attributes does not relate to the presence or absence of other attributes. This method predicts future instances based on exploited data (Seeja, et al, 2014).

C. Random Forest

A group of decision trees is known as a random forest. Like binding strands of yarn into a rope, individual trees make the forest. While weak on their own, the trees are strong as a group. The group works together to classify incoming objects and detect fraudulent transactions.

Income objects go through each tree for classification. The trees compare classification results and decide on a final classification for each object based on the group observations. This is a fast-moving method that can handle a lot of incoming information (Seeja, et al, 2014). The disadvantage to this method is that each transaction must be run through the whole system, presenting a lot of information to consume (Abdou, et al, 2009).

D. Geo Location

Geo location can be used to identify if a user’s information is being impersonated. Every user has an IP address associated with their specific computer. Neural networks can use this information to make sure a user’s IP address matches up when shopping online. The neural network can compare the IP being used to the IP on record for the last year or two, as well as determine if a proxy server is being used to hide the IP address (Seeja, et al, 2014). This can also be used in the case of physical card use, as a purchase a long distance from a user’s physical address could signal a fraudulent transaction.

E. Data Clustering

In Data Clustering, customer accounts are clustered together in groups for observation purposes. By observing comparative data between these customer accounts under normal conditions, a neural network can determine when one of these customer accounts is being misused. Suspicious activity can be flagged, and fraud analysts will investigate any alarms raised (Abdou, et al, 2009). This method requires time for investigation, however.

F. Hidden Markov Model

The Hidden Markov Model is trained in the normal behavior of a customer. Customers are classified as having high, middle, or low expenditures. If an expenditure does not fit in the classification, it is labeled as possibly fraudulent. Transaction logs help the Hidden Markov Model determine if an account is behaving properly (Patell, et al, 2015).

G. Genetic Algorithms

Genetic algorithms differ from static algorithms. A genetic algorithm learns as time goes on. The process is repeated through multiple generations to refine the algorithm. Better results are seen as the algorithm progress, continually increasing detection rates (Pushpalatha, et al, 2017).

H. Communal Data

Communal Data uses the approach of similar information flagging credit card applications. Personal information is ranked in priority. Factors such as social security number, name, and address are looked at in this method. If two applications come in with similar data but slight changes, the method knows to consider that one may be fraudulent and looks for discrepancies to detect which one (Arthisree, et al, 2013).

V. OTHER FRAUD DETECTION METHODS

Besides data mining techniques, other methods can be used to ascertain if a credit application or use of card is a legitimate transaction. Certain personal information must be changed when a fraudster is using a card for a purchase. These factors include email, phone number, and shipping address and bio scanning. Fraud detection scientists are also a valued factor in interpreting data from various applications.

A. Email and Phone Number

Credit card customers are required to use their phone number and email when applying for credit. Users can elect to receive emails about purchases made on their card or can use their information to authenticate their identity when checking their bill online. Being able to instantly access recent transactions, instead of waiting for a statement in the mail, means a user can quickly ascertain if a fraudulent transaction has occurred. This also means that a fraudulent user must use a different email or phone number when signing up for a card, which a neural network can be trained to look for (Seeja, et al, 2014).

B. Shipping Address

Just as email and phone numbers must be different for fraudulent users, a fraudulent user cannot ship goods to the real customer’s house. Neural networks can be trained to look for a change in address. The only issue here is that some customers do send gifts to different addresses, which may result in a false alarm or data that is difficult to interpret correctly in a preventative manner (Seeja, et al, 2014).

C. Bio Scanning

Another method of protecting users from fraudulent transactions is to require bio scanning. Many smartphones today are enabled with finger print scanners. These scanners are currently used for unlocking phones and accessing certain applications, such as online banking. If a user is required to scan their fingerprint when making an online purchase, identity can be verified at the time of the transaction (Priyadharshini, et al, 2012). 

D. Fraud Data Scientists

Fraud Data Scientists are needed to develop and monitor various applications of fraud detection. These scientists need skills in qualitative reasoning. Also, due to different fraud detection methods, it is important for a fraud data scientist to have a background in programming. He or she also needs to have an understanding of business, as well as the ability to communicate clearly and effectively with others. Creativity is also a desired trait, as this will help in effective brainstorming and problem-solving (Beasens, et al, 2015).

VI. CONCLUSION

Credit card fraud is on the rise in the online world. With current methods, it is easy to steal personal information and the creative minds of con artists are constantly inventing new ways to avoid detection while stealing information. Data mining has been useful in allowing for the development of applications to thwart these efforts. 

These applications must be continually examined and improved to stay ahead of the efforts of fraudsters. A combination of applications is best practice for any business. Utilizing more than one application in the detection of fraudulent transactions will greatly reduce the occurrence of said transactions. Human resources can provide back up for automated systems and investigate claims made by applications.

REFERENCES

[1] R. Gayathri and A. Malathi. “Investigation of Data Mining Techniques in Fraud Detection: Credit Card,” International Journal of Computer Applications, Vol. 82, No. 9, November 2013.

[2] F. Ogwueleka. “Data Mining Application in Credit Card Fraud Detection System,” Journal of Engineering Science and Technology, Vol. 6, No. 3, pp. 311-322, 2011.

[3] T. Razak and G. Ahmed. “Detecting Credit Card Fraud using Data Mining Techniques - Meta-Learning,” Indian Journal of Science and Technology, Vol. 8, October 2015.

[4] S. Kumari and A. Chouby. “A Review on Various Techniques and Approaches for Credit Card Fraud Detection,” International Journal of Scientific Research Engineering & Technology, Vol. 6, Issue 5, May 2017.

[5] K. Seeja and M. Zareapoor. “FraudMiner: A Novel Credit Card Fraud Detection Model Based on Frequent Itemset Mining,” The Scientific World Journal, 2014.

[6] J. Akhilomen. “Data Mining Application for Cyber Credit-card Fraud Detection System,” Proceedings of the World Congress on Engineering, Vol 3, July 3-5 2013.

[7] A. Desai and R. Desmukh. “Data mining techniques for Fraud Detection,” International Journal of Computer Science and Information Technologies, Vol. 4, 2013.

[8] C. Phua, V. Lee, K. Smith, and R. Gayler. “A Comprehensive Survey of Data Mining-based Fraud Detection Research,” Monash University.

[9] V..Priyadharshini, and G. Macriga. “An Efficient Data Mining for Credit Card Fraud Detection using Finger Print Recognition,” International Journal of Advanced Computer Research, Vol. 2, No. 4, Issue 7, December 2012.

[10] M. Fahmi, A. Hamdy, and K. Nagati. “Engineering & Informatics Data Mining Techniques for Credit Card Fraud Detection: Empirical Study,” Sustainable Vital Technologies, November 2016.

[11] M. Albashrawi. “Detecting Financial Fraud Using Data Mining Techniques: A Decade Review from 2004 to 2015,” Journal of Data Science, pp 553-570, 2016.

[12] S. Soltanitziba and M. Balafar. “The Study of Fraud Detection in Financial and Credit Institutions with Real Data,” Computer Science and Engineering, 2015.

[13] D. Abdou and J. Pointon. “Credit Card Fraud and Detection 

Techniques: A Review,” University of Salford Manchester, 2009.

[14] T. Hisar and M. Bansal. “Survey Paper on Credit Card Fraud Detection,” International Journal of Advanced Research in Computer Engineering & Technology, Vol. 3, Issue 3, March 2013.

[15] N. Malini and Dr. M. Pushpa. “Analysis on Credit Card Fraud Detection Techniques by Data Mining and Big Data Approach,” International Journal of Research in Computer Applications and Robotics, Vol. 5, Issue 5, May 2017.

[16] S. Redi. “The Application of Data Mining Technology,” University of Applied Sciences, 2016.

[17] S. Patell and P. Desai. “A Survey of Online Credit Card Fraud Detection using Data Mining Techniques,” International Journal for Scientific Research & Development, Vol. 3, Issue 10, 2015.

[18] B. Baesens, V. Vlasselaer, and W. Verbeke. Fraud Analytics Using Descriptive, Predictive, and Social Network Techniques: A Guide to Data Science for Fraud Detection. SAS Institute Inc, Cary, North Carolina, 2015. 

[19] B. Pushpalatha and C. Joseph. “Credit Card Fraud Detection Based on the Transaction by Using Data mining Techniques,” International Journal of Innovative Research in Computer and Communication Engineering, Vol. 5, Issue 2, February 2017.

[20] K.S.Arthisree and A. Jaganrai. “Identify Crime Detection Using Data Mining Techniques,” International Journal of Advanced Research in Computer Science and Software Engineering Research, Vol. 3, Issue 8, August 2013.