As we climb up the ladder of digitization, we get more comfortable with the idea of mobile payments. While technology evolves in one direction, the fraudsters seem to come up with innovative ways to defraud the customer. As mobile payments become easy, there has been a rising number of mobile payment frauds. This has become one of the major concerns of mobile payment products like UPI.
Digital payments have seen exponential growth over the last decade. According to RBI, digital payments – NEFT, IMPS, UPI, etc. have shown a remarkable growth of about 55% and 43% in terms of volume and value, respectively over the past ten years. In March 2021 alone, UPI transactions in India crossed 2.7mn in volume and USD 68bn in transaction value. However, with this, we also witnessed a sharp rise in digital wallets and mobile payment frauds. Data from Mumbai Cyber Cell showed a growth of over 70% last year in mobile payment frauds. In order to mitigate the risk of mobile & online payment frauds, regulations have been issued, which include the need to implement fraud management systems in mobile payments.
Amidst the Covid-19 pandemic, Cyber crime witnessed a rise of 300% in the country reaching around 1,158,208 in 2020 compared to 394,499 in 2019, according to Union home ministry, citing the data from CERT-in, Computer Emergency Response Team.
Fraud risk management solutions have been driven by rules, which require human analysts to add new rules every time a fraudster innovates. A rule based system operates on algorithms that are manually written by analysts. Fraudsters innovate at a faster pace and therefore the need to create rules automatically using machine learning, is the key to reduce and mitigate frauds.
Machine Learning in Fraud Detection
Machine Learning can prove to be a boon for fraud management systems. It has the potential to learn new patterns based on data and also identify outliers or anomalies, which require human analysis. While machine learning models can be devised to learn from large data sets, to be effective in preventing fraudulent transactions they need to operate at low latency, less than 30 milliseconds. Drona Pay has built machine learning models that help banks, fintechs, and payment processors in realtime decisioning.
Drona Pay is not only an upgrade to the rule based system but a complete rewrite, given that most fraud systems were not built for high throughput and low latency. Drona Pay leverages a mix of supervised learning, unsupervised learning, natural language processing and graph models to identify fraudulent transactions. Several ML algorithms have been evaluated to identify suitable models. Drona Pay has identified models that demonstrate accuracy, explainability and offer realtime response.
Machine Learning vs. Rule based systems in fraud detection
A rule based fraud detection system grows unwieldy over time as rules keep on getting enhanced and added into the system. The maintainability of rule based systems reduces over time as every new rule tries to make it to the top. ML based fraud detection systems operate in a multi-dimensional space, at high speed and measurable accuracy. The models offer a range of levers through which performance can be tuned. While the rule based approach requires manual work to create new rules every time, ML models automatically detect new frauds based on pattern matching and outlier detection, which helps them cover a larger set of emerging fraud scenarios. ML models improve the user experience by reducing friction through the reduction in the number of verifications and steps ups.
To learn more about how to secure your transactions using Machine Learning, while enhancing customer experience, visit us at www.dronapay.com
Comments