Data mining is the extraction of hidden trends and patterns in a huge database to convert the data into knowledge. This knowledge helps in predicting future trends or patterns which enables companies to take data driven proactive decisions to increase revenue or reduce cost or both. It uses various automated analytical tools to discover relationships and trends in data which may help in predicting the future behaviour. In case if there are any deviations from the predicted behaviour of the consumer, necessary action can be taken.
For example, if a system can determine or identify based on historical data that which accredit card accounts would go delinquent this month, the company can send those account holders some reminders to make the payments and avoid going delinquent. This is done by analysing the historical data and understanding the behaviour of the consumer. Identify these trends is almost impossible by traditional data analysis methods. Also, when the number if data points are in millions, it would take ages for a human to analyse the same data with high probability of missing the hidden trends.
Data mining in relation with credit card fraud prevention
When a transaction is being made using a credit card, if a system can predict whether the transaction is genuine or fraud, then credit cards frauds can be controlled in a big way. For example, consider a credit card user who has always made transactions using a credit is the city of London only. When a credit card is swiped on a terminal or online, it is first verified and only then the funds are transferred to the merchant’s account. For the user above, the credit card system suddenly detects the credit card of the user is swiped at a merchant’s shop in Ireland during the verification process. In such a scenario, either the transaction can be declined or some additional personal information be asked to the user like first school, mother’s maiden name etc. In case if the questions are answered incorrectly, the transaction can be declined. In a similar manner if any suspicious transactions are detected by the credit card processing system, the transaction can be declined thus preventing the fraud from happening. The question is how dos the system define “suspicious” transaction. In the example above, it was a little easy to identify the location however what happens when the number of credit cards issued by a bank are in millions and each customer has different information, different location, and different shopping habits? The data cannot be easily analysed and the future behaviour cannot be predicted easily.
This is where data mining can help in solving the problem to a great extent. Data mining involves analysing large amount of data and identifying hidden trends which cannot be identified by simple data analysis. Also, by the various models in data mining, the future behaviour of a user can also be predicted and if there is any mismatch in it, wither the transaction can be declined or some additional confidential information be asked to the user.
Many banks and credit card issuing companies have started using data mining techniques in fraud detection. However it involves lot of cost and resources to be fully implemented. Also, designing and developing models for credit card fraud detection is also a tedious task. However many companies are relying on it to prevent the loss of millions of dollars suffered every year by them, merchants and credit cad users.
Some Data mining concepts/models in credit card fraud detection
1. Neural data mining
Neural networks establish the relationship between a dependent and independent variable and then predict the dependent variable. This is a very complex process however the most commonly used data mining technique in credit card fraud detection. For example, the neural data mining can establish the relationship between a credit card user and his shopping habits and if there is any transaction is made using the credit card outside the habits of the credit card user, the system can either decline the transaction or probe questions to the user which only he is expected to know.
Although the model sounds very easy, it is very difficult to establish the relationship between the dependent and independent variable especially when the user database is in millions and with very dynamic independent variables. Also the accuracy of the model is based on the confidence level and probability and neural networks do not guarantee the fraud detection.
2. Decision Trees
Decision trees are the most simple of all the data mining techniques used in credit card fraud detections. By using the data mining technique the credit card company can classify its credit card holders into various categories based on certain rules or series of rules. For example, a credit card company can classify its customers as business travellers/frequent travellers or customers limited to few city or towns. If the credit card of a consumer classified as in a class which never travels or rarely travels is used outside the city limits or some far distant location, the transaction can be easily declined or some affirmative action ca be taken to avoid the potential fraud. Another example is if a customer has not submitted his passport number at the time of account opening or has not ‘updated the passport number in the bank records, is found to be shopping at an international location can be detected using the decision tree credit card fraud detection system.
However, the challenge faced in this technique is that the rules for the credit card fraud detection system have to be frequently modified with time. It does not hold as a one time solution and managing the same becomes very difficult for any credit card company.
Also, the rules cannot be defined for each and every type of transaction as in this world of technology the credit card frauds can take place from any location.
3. Hidden Markov model
Hidden Markov model is a model designed using statistics. In this the state of an independent variable is not known based on its impact on the dependent variable, its state can be determined. Hidden Markov models works on the fundamentals of probability. A hidden Markov model is designed using the normal user behaviour. If any transaction made on the credit card of a user shows a low score on the hidden Markov model probability score, it can be assumed as a fraudulent transaction and declined. However, the limitation of this model being that low probability does not always mean fraud and high probability does not always mean a genuine transaction. Foe example, a user buys cigars once in a week from the same shop; the transaction would always be considered genuine by the hidden Markov model whereas it may be a fraudulent transaction.
Conclusion
The credit card market is increasing on a daily basis and so are the credit card frauds. There is an urgent need to prevent these frauds in order to safe guard the interests of users, merchants and banks. Data mining has the ability to minimize the frauds however data mining is still too young to for the banks and credit card issuers to rely on it. However, most companies are testing various data mining models to develop the ultimate fraud detection system.
| < Prev |
|---|
