The future of mobile banking is clear. People love their mobile devices and banks are making big investments to enhance their apps with digital features and capabilities. As mobile banking grows, so does the one aspect about it that can be wrenching for customers and banks, mobile device fraud.
Problem: To implement near real-time fraud detection
Most mobile fraud occurs through a compromise called a SIM swap attack in which a mobile number is hacked. The phone number is cloned and the criminal receives all the text messages and calls sent to the victim’s mobile device. Then login credentials are obtained through social engineering, phishing, vishing, or an infected downloaded app. With this information, the criminal can impersonate a bank customer, register for mobile access, and immediately start to request fund transfers and withdrawals.
Artificial Intelligence (AI) models have the potential to dramatically improve fraud detection rates and detection times. One approach is described in the Mobile bank fraud solution guide. It’s a behavioral-based AI approach and can be much more responsive to changing fraud patterns than rules-based or other approaches.
The solution: A pipeline that detects fraud in less than two seconds
Latency and response times are critical in a fraud detection solution. The time it takes a bank to react to a fraudulent transaction translates directly to how much financial loss can be prevented. The sooner the detection takes place, the less the financial loss.
To be effective, detection needs to occur in less than two seconds. This means less than two seconds to process an incoming mobile activity, build a behavioral profile, evaluate the transaction for fraud, and determine if an action needs to be taken. The approach described in this solution is based on:
- Feature engineering to create customer and account profiles.
- Azure Machine Learning to create a fraud classification model.
- Azure PaaS services for real-time event processing and end-to-end workflow.
The architecture: Azure Functions, Azure SQL, and Azure Machine Learning
Most steps in the event processing pipeline start with a call to Azure Functions because functions are serverless, easily scaled out, and can be scheduled.
The power of data in this solution comes from mobile messages that are standardized, joined, and aggregated with historical data to create behavior profiles. This is done using the in-memory technologies in Azure SQL.
Training of a fraud classifier is done with Azure Machine Learning Studio (AML Studio) and custom R code to create account level metrics.
Recommended next steps
Read the Mobile bank fraud solution guide to learn details on the architecture of the solution. The guide explains the logic and concepts and gets you to the next stage in implementing a mobile bank fraud detection solution. We hope you find this helpful and we welcome your feedback.
