Advancements in computing technologies and the expanding use of e-commerce platforms have dramatically increased the risk of fraud for financial services companies and their customers. Failing to properly identify and prevent fraud is an expensive proposition that costs the financial industry billions of dollars per year. In fact, a recent Nilson Report estimates that worldwide losses due to credit card fraud alone would top $27.69 billion in 2017, increasing 11 percent over the previous year.
As the annual costs of fraud rise and fraudsters continually adapt their tactics to avoid detection, financial companies are doing everything they can to better protect themselves. Traditional approaches to fraud detection were rule-based, employing a set of logic statements to continually query transactions and flag suspicious activity for human review. However, there are issues emerging with these methods, primarily that the rules can be so generalized that they cause millions of legitimate transactions to be turned away or declined. This approach is also not as effective at detecting and exposing new fraudulent techniques in real-time.
This is causing a number of artificial intelligence (AI) techniques, including deep learning, to become increasingly popular and effective methods of fraud detection. Deep learning algorithms can quickly sift through and analyze vast quantities of data and uncover transactional anomalies or suspicious patterns. These algorithms are built on artificial neural networks that are comprised of many layers and modeled after the human brain. They are designed to uncover patterns in tremendously large datasets and independently learn new concepts from raw data.
Deep learning algorithms can be supervised, partially supervised, or unsupervised. In fraud detection, unsupervised models can be used to isolate anomalous activity, while supervised models can be used to distinguish which anomalies are actually fraudulent and which are just unusual. This allows financial companies to allow legitimate transactions to continue processing while they also identify and flag potentially fraudulent activity.
Neural networks have been used for fraud detection for decades, but the high performance computing (HPC) technologies and large data volumes available today have dramatically improved the effectiveness of these techniques. Here are a few examples of how today’s companies are leveraging deep learning for fraud detection and prevention:
- By aggregating structured and unstructured data from historical databases, claims management systems, and even social media platforms, insurance companies can compare the characteristics of new claims against those of past losses in order to flag potentially fraudulent activity.
- E-commerce giant PayPal uses a combination of three types of machine learning algorithms – linear, neural network, and deep learning – in order to quickly distinguish trustworthy customers and put them in the “express lane” to a complete a transaction.
- Algorithms can be trained to flag credit card transactions that appear suspicious based on logical rules, such as a card being used in an unusual region or time of day, repetitive transactions, or if verifying card details was attempted multiple times.
Deep learning is a highly complicated exercise that requires tools capable of delivering extreme compute power and exponential scaling. Hewlett Packard Enterprise (HPE) offers a comprehensive, purpose-built portfolio of computing innovations to accelerate deep learning applications and insights across the enterprise. Our industry-leading platforms and supporting services and expertise can help financial services organizations succeed with deep learning algorithms, and ultimately, better protect themselves and their customers in this highly digital age.
For more news and case studies regarding the use of deep learning for fraud detection, please follow me on Twitter at @pango. And for general discussion about AI, deep learning, and HPC, please check out the HPE HPC Twitter page at @HPE_HPC.