This post was originally published on Global Banking & Finance.
Fraud detection is a billion-dollar problem in finance, affecting consumers and banks alike. Deploying and running advanced analytics on data as it’s born will help financial institutions detect and prevent money laundering and other fraudulent activities while eliminating the need to provision new data stores for fraud detection workflows. The real-time insights gleaned trigger on-the-spot workflows to protect against live fraud at the time of the transaction.
Leading banks and capital markets firms have begun to leverage real-time analytics and machine learning for many initiatives; including risk management, fraud detection, compliance and consumer metrics, to gain a competitive edge and comply with regulations that are required for extreme performance on big data workflows.
Machine learning is getting better at identifying potential cases of fraud across many different fields. It is being used to fight money laundering, and companies like PayPal are building tools that compare millions of transactions at millisecond speeds to precisely distinguish between legitimate and fraudulent transactions on the buyer and seller sides.
Operationalizing Machine Learning for Finance
Taking the leap from data science to operationalizing machine learning requires successful integration with business applications. No matter how clean and organized data is, or how sophisticated statistical models are – if the last mile of application integration is not possible, then financial institutes can never fully leverage the benefits of machine learning.
For example, if a financial transaction has been identified as fraudulent, this insight can only be turned into action through this essential integration. The right organizational stakeholders need to be notified, and the relevant workflow needs to be executed. Otherwise, the fraudulent attempt will go unchecked.
Detecting Fraud and Money Laundering with In-Memory Computing (IMC)
Unlike traditional databases, an in-memory computing can handle real-time massive workloads and processing tasks at millisecond speeds. And smart integration with data lakes, storing multi-petabytes of data, simplifies access to historical data. This powers faster and smarter machine learning insights, providing a simpler, faster workflow.
Fraud Detection in Real-time: A Financial Use Case
A leading payment solution provider has leveraged GigaSpaces In-memory Computing Platform to provide tools that facilitate a secure exchange between financial organizations and offer solutions for responding instantly to evolving fraud challenges. Such customers exchange information and knowledge to obtain a single view of fraud activity across the enterprise and manage fraud on a cross-institution basis.
Fraud challenges exist for real-time mobile payment applications as well as detecting check fraud when there are simultaneous check deposits at different banks In order to address these, the core platform technology not only ingests 4 TB daily but also handles 1.5M events per second, with a response time of milliseconds. This gives banks and financial institutions the processing power to handle all of this insight and validate it against large datasets of both live and archived data, correlating the immediate transaction with the historical behavior of the specific user profile. They otherwise wouldn’t be able to with traditional RAM.
Business Challenge
- Detecting fraud on mobile payment applications in real-time
- Detecting the deposit of the same check in multiple accounts at different banks in real-time
- User experience: application availability 24×7
- TCO reduction: reduce dependency on expensive RDBMS (Oracle)
Technical Solution
- IMC Platform to ingest 4 TB of data daily
- Fully consistent transactional In-Memory Map-Reduce
- Millisecond response
- Analyze and validate against a large dataset of live (multiple TB) in memory and archived data (to Cassandra NoSQL and Hadoop)
Results
- Sub-second response for accurate fraud detection to stop the transaction
- TCO Reduction: RAM and SSD for runtime data compared to Oracle DB or SAN
- Fault-tolerant, highly available, scaling on demand
If a faulty transaction is detected, it can be rejected instantaneously. While this might sound like it puts a burden on the user experience, it doesn’t. Transactions are completed in seconds as the actual processing and analysis take on millisecond. On the enterprise side, it is built for on-demand scaling and mission-critical availability with proven zero downtime
Final Thoughts
Real-time analytics and machine learning will enable companies to actively prevent transactional fraud. Reacting after the fact is too late and has a negative impact on both costs and customer experience. In-memory-computing allows banks, credit unions and larger banks alike to act in the moment, rather than getting burned in the future.
GigaSpaces’ InsightEdge Platform contains all the necessary SQL, Spark, streaming, and deep learning toolkits for scalable real-time data-driven solutions. TCO is optimized with multi-tiered storage across RAM, SSD, and Storage-Class Memory (3DXPoint).
Architectural simplicity is supported via unified management, security, and orchestration on the cloud, on-premise, or in a hybrid model.
Designed specifically to support high-throughput transactional applications for financial, insurance, retail/e-commerce, telecommunication, and IOT scenarios, GigaSpaces InsightEdge Platform is powering the insight-driven organization by operationalizing and maximizing the full potential of machine learning.
To learn more about how Machine Learning and our InsightEdge Platform can bring value to organizations in many industries read our latest eBook on Machine Learning.
