(Credit: Mindtree) -What is machine learning?

Machine learning in simpler terms is how machine (software) continuously learns and improves itself about a subject as the amount of data it process keeps increasing. Technically, Machine learning is self-learning algorithms which are prewritten or predefined adhering to some specific rules or factors. As more and more data is processed by the machine, the algorithms improve leading to better predictions.

Why machine learning in payment transactions?


Majority of people nowadays use credit card, debit card or mobile/digital wallets for their online and/or offline purchases for the convenience and simplicity they offer. But with increase in volume of such transactions, there is also increase in the fraud which are happening in these transaction. Also, since there are multiple channels such as online purchases, bill payments and retail purchases through cards, and wallet payments etc., chances of fraud also increases.

These frauds are causing huge amount of losses to both the customers as well as to the banks. On customer front, they have to go through lot of hassle in blocking the card, reporting the stolen card etc., whereas on the bank front, they have to bear all or most of the loss on these fraudulent transactions along with impact on its brand image. According to Nelsen report, fraud losses incurred by the banks and merchants were about $16.31 billion in 2014.Fraud grew by 19% outpacing volume transaction which grew by 15%.

So it imperative for banks, FIs and credit card companies to have implement mechanisms to detect and prevent any type of fraud and in real time. Machine learning technology for fraud detection seem to be a compelling fit. It makes a perfect case for card and mobile wallet transactions. Voluminous data of transactions are available with banks, and they keep growing. Machine learning, when applied will help predict future pattern and alerts for any abnormality in the pattern.

Factors which can be considered in building rules/algorithms



  1. High transaction amounts: Suppose an average spend of a customer in the last six months is around $2000, and there happens a transaction which is way higher than this. The bank can immediately react by automating a second level of authentication. They can trigger a call or message and request the customer to verify the transaction. So if the customer hasn’t initiated this particular transaction they will deny initiating such transaction and hence the system will terminate the transaction this prevents misuse or fraud, if any. If customer

  2. Uncustomary transaction amounts: Suppose that the bank notices a transaction that is way higher than the maximum amount of transaction. The bank can immediately react by automating a second level of authentication. They can trigger a call or message and request the customer to verify the transaction. This prevents misuse or fraud, if any.

  3. Transaction in a new geography: If the bank detects a transaction that happens outside the residence of the customer, it can react efficiently. As example, A person residing in New York mostly does all his transaction in and around this area. However, if there is a transaction that occurs in a someplace in Middle East or Africa, the bank system can take preventive measures to cancel the transaction.

  4. Increased frequency of transactions: Suppose that the bank notices that the number of transactions is way higher than customary ones, the bank can immediately react by automating a second level of authentication after the average number.

  5. Unusual time of Transaction: Suppose a customer normally transacts between 10.00 am to 7.00 pm, but there is any transaction occurring in late night say 11.00 pm or early morning 6am, this could be an indication for a fraud and accordingly alert will be triggered.


The above mentioned use cases are just examples and there are many more factors to consider. A combination of factors are used in defining the logic for algorithms in order to prevent fraud in real time to ensure both genuine and fraudulent cases are handled properly and avoid customer dissatisfaction.

Leverage Machine Learning Use Case Infographic



Suppose a customer swipes his credit card at a POS. Once the card is swiped, algorithms based on the some of the factors discussed above and many others comes into picture. System interact with the already collected data and do the validation for the authenticity of the transaction. If it is seen that there is abnormality or deviation from the pattern build for that particular person for e.g. his maximum single card transaction over many years is $1000, but this transaction is for $5000, the system will trigger an alert and it will ask the customer either through message or call for validating this transaction. If customer confirm that he is the one who has done this transaction then the transaction becomes successful or else the system stops further process of the transaction. All this happens real time and within fraction of time.

Conclusion


With banks and credit card companies facing issue of fraudulent transaction, be it card transactions or online transactions, it becomes imperative for banks to have real implementation rather than just the trial version implementation of machine learning based algorithms. This technology can help a great deal to banks at least in reducing the number of fraudulent transactions and thereby reducing the financial losses.


- Karan Malhotra -




Note:  Karan Malhotra is Associate Consultant, Banking and Financial Services COE.


Karan.Malhotra@mindtree.com
저작권자 © KIPOST(키포스트) 무단전재 및 재배포 금지