Despite the introduction of more secure chip technology, credit card fraud is still one of the biggest concerns of banks and credit card companies. After all, fraudsters continually change their tactics to get around any security measures that are put in place. A recent Nilson report projects that global total loss due to credit card scams and fraud will be a whopping $32.96 billion by 2021. Many financial companies are now looking at predictive analysis to see if they can depress these numbers.
Beyond Pattern Recognition Software
Artificial Intelligence has been at work for quite a while at differing levels in bank and credit card companies. AI development has only gotten better since online transactions started to turn up the volume. Many of us would have received a telephone call, a text or email alert when an unusual transaction was detected. This was the first time that pattern recognition software came to the forefront. The usual general rules of fraudulent behavior were set aside and it was found to be much more effective to create patterns for each account. Pattern recognition software or neural networks have looked at patterns in individual accounts and cards to know when something seems out of tune. Automated notifications are sent out to the customer and, in many cases, transactions are stopped.
In recent times, predictive analytics has taken this much further and found a wide adoption among banks and credit card companies in particular.
Using Big Data to Successfully Fight Credit Card Frauds
Technology versus credit card fraud will be a continuing battle, but Big Data and automated notifications are helping to give more ammunition to the ‘Good Guys’. It is a known fact that while scammers keep trying to beat the system in new ways, their new threats will include some pieces of their old methods. Predictive analytics can help to build solutions that can identify these ‘fraud signatures’ before they can even take hold and make lasting damage. This machine learning reacts continuously to prevent recurrence of such threats with a high degree of accuracy.
Fraud risk management analytics aims to build an estimation model to distinguish the Highest Possible True Positive Rate (TPR) and the Lowest Possible False Positive Rate (FPR). This forms the basis for creating the Estimation Model to accurately forecast and monitor frauds. The problem lies in getting the TPR and FPR right, so refining the bank data is the first step. Then, companies can train their Estimation Model using oversampling methods. After all, no customer wants their card declined when they are at a store or on holiday and making a legitimate purchase. New machine learning techniques have been developed that reduce these false positives to a great extent.
Predictive Analytics Can Give the Go-ahead on Each Transaction
The estimation models have been built by researchers using ginormous data sets. Think 900 million transactions from about 7 million individual cards. Out of this, about 120,000 were known as fraudulent transactions. These researchers have used subsets of this data to test their model. Patterns of transactions, as mentioned earlier, continue to be what the predictive models are based on. In the machine learning period of training, different highly customized variables are created based on looking ‘deeply’ at each and every transaction.
This can be more easily explained with a real-life transaction history. A customer is found to make 2 or 3 online purchases every month, and the amount spent never exceeds $500. This looks like a person who loves shopping and discount offers, so a variable will be created for this category. The machine then creates an if/then decision tree with features that will or will not point to fraud. Now, when a real-time transaction is run through this decision tree, the predictive model will decide then and there whether the transaction is fraudulent or not.
Today, federal law places the liability on the account holder for a maximum of $50, no matter the fraudulent transaction amount. Several credit card companies have voluntarily adopted a zero liability policy for their account holders. These policies make it even more vital for adoption of the best predictive analytic models that are available.