Machine Learning in Real-Time vs Rules-Based Detection
ML focuses on a streamlined and simplified customer-oriented protection system that enhances or completely changes processes.
Join the DZone community and get the full member experience.Join For Free
Unforeseen Levels of Increased Online Purchasing Leads to Higher Payment Fraud
The unparalleled surge in online purchasing during COVID-19 has led to a 40% rise in online retail fraud attempts, according to the New York-based Fraud.net. Payment fraud had, anyway, been increasing in recent years, with the rise of e-commerce. For instance, online shopping fraud grew by 30% in 2017 – twice as fast as e-commerce sales. A recent True Cost of Fraud Study by LexisNexis found that online credit card fraud is estimated to reach $32 billion in 2020.
Moreover, the pandemic has pushed online shopping to a whole new level.
New Payment Methods Add Complexity and Vulnerability to Transactions
Thus, online retailers realize that their current strategy of managing fraud through rule-based detection fails because of crucial changes in the dynamics of financial transactions today.
Customarily, fraud detection models that banks developed were for physical credit card transactions. Today, online purchases do not require the physical handling of a credit card. There is an influx of newer payment methods such as Debit Cards, Electronic Funds Transfer (EFT) or Instant EFT, Virtual Credit Cards, Prepaid Cards, Prepaid Vouchers, Merchant Accounts like PayPal, Use Store Credit and Cryptocurrencies, to name some, where funds can be accessed directly from a bank account, and, therefore, are not protected by the existing rules-based protection system.
For instance, the rules-based system is based on identifying a known fraud pattern that determines if a transaction is authentic or fraudulent. However, its rigidity promotes inaccuracy; benign transactions get flagged as fraudulent and are declined, while fraudulent transactions are accepted as legitimate.
According to recent studies, about 15% of all Card Not Present (CNP) transactions are false positives, which lead to an annual loss of revenue of US$ 118 billion to online retailers. Worse still, over 1/3 of wrongly declined customers never return. Thus, the fear of fraud may be more damaging than fraud itself. On the other hand, if fraudulent transactions are missed, retailers can lose significant revenue.
Thus, inadequate protection makes new transactions vulnerable, requiring a different kind of fraud model for speedier and more secure transactions.
From Rules-Based Protection to Machine Learning (ML) Models
In other words, e-commerce requires the real-time protection of Machine Learning (ML) models of Artificial Intelligence (AI); ML focuses on a streamlined and simplified customer-oriented protection system that enhances or completely replaces faulty, tedious, and slow manual processes with risk analytics, that will also lead to high customer satisfaction.
ML methods also benefit from the opportunity to learn from different variables. It follows that a limited number of variables will diminish the advantage ML methods have over rules-based models, leading to an algorithm that cannot identify any meaningful patterns. Having access to many records for model development is especially important for ML methods, as fraudulent transactions happen in the range of 1 in 1,000 to 1 in 50,000.
Rule-Based vs. Machine Learning (ML)
So, it is vital to understand ML offers online retailers a new opportunity to be proactive, where rule-based methods are essentially reactive. The reactivity is what held back financial institutions and retailers from effectively and efficiently preventing fraud. ML enables real-time insights, identifies irregularities, and detects subtle changes in large chunks of dynamic data sets.
While rules-based systems rely on pre-programmed rules to spot behavioral changes or predict outcomes, rapidly evolving ML systems require a more flexible approach to identifying fraud.
Advantages of ML
- Rules-based is time-intensive; ML facilitates real-time processing.
- Rules-based requires manual work and supervision; ML enables automatic detection of anomalies.
- Rules-based requires multiple steps for verification, impeding user experience; ML reduces, and simplifies verification.
- Rules-based identifies obvious fraud patterns; ML can discover hidden fraud by focusing on subtle pattern changes.
- ML relies on algorithms that become more efficient, effective, and economical as data sets increase. The rules-based model becomes more expensive, with more data.
Applying ML in Financial Services
ML can easily prevent online credit card fraud by using technology to identify and stop high-risk orders in real-time. Moreover, tying ML to other emerging technologies like biometrics enables simplified verification and authorization, reduces fraud, and improves customer experience.
Insurance and insurtech can also benefit from ML models as they can detect irregularities, flag them for review, and enable more accurate and efficient manual reviews.
Currently, all branches of insurance lose about US$ 80 billion annually to fraud.
The Danger of Fraudsters Using AI on Banks
However, in using AI for fraud detection, financial institutions need to be keenly aware that fraudsters, in turn, can employ AI against banks. They may not regularly stage cyberattacks and steal money or make fraudulent payments, but they could distort ML processes to alter the way AI engines work.
Steve Holt, Partner at Ernst and Young, said, “One of the big concerns, especially at the regulatory level for the future, is ultimately the underlying data integrity. If the attackers don’t do big enormous payouts immediately but attempt to alter the underlying data, how would that be spotted?”
That is indeed a red flag for banks to be on high alert for possible criminal manipulation of ML processes.
However, as the famed Marie Curie once said, “Nothing in life is to be feared, it is only to be understood. Now is the time to understand more, so that we may fear less.”
Opinions expressed by DZone contributors are their own.