Financial institutions everywhere are ramping up cybersecurity efforts against the ongoing issue of bank fraud. This is especially the case with PayPal, as their efforts towards utilizing deep learning against fraud, and the company is using homegrown artificial intelligence engine built with open-source tools. PayPal Deep Learning Methods Against Fraud
The company’s reputation as a payment vendor is always at high stake due to its high visibility and massive payment volumes. PayPal generates $10,900 in payments every second, and it handled 4.9 billion payments in 2015 for 188 million customers in 202 countries totalling $235 billion in sales. Fraud is always possible via theft of consumer data in breaches such as phishing campaigns that trick users into entering their credentials. With such cases, PayPal relies on intensive, real-time analysis of transactions.
Deep learning utilizes artificial neural network algorithms that can effectively gather data insights and recognize patterns. There are now many deep learning applications in image/speech recognition, and language analysis. Now these systems are turning out to be effective in recognizing the patterns and characteristics of fraudulent activities against fintech companies. While PayPal has been experimenting with machine learning-based solutions for the past decade, they have already transitioned from machine-learning-based pattern recognition to deep learning techniques.
Most of PayPal’s data is stored in Hadoop as a data platform to handle the constantly growing data volumes and varieties of data, allowing PayPal’s data scientists more flexibility in handling the data. As further detailed in their engineering blog, the raw clickstream data is processed in Hadoop through a cleaning phase. PayPal uses semi-structured data in Hadoop, for predetermined business intelligence and big data analytics projects and stores it in the cloud for all PayPal employees to access globally. It collects more than 20TB of log data every day for sentiment analysis, event analytics, customer segmentation, recommendation engine and sending out real-time location based offers.
Ultimately, deep learning has been effective in analyzing factors such as timelines, location, etc. as part of payment transactions. Several kinds of algorithms analyze thousands of data points in real-time, such as IP address, buying history, recent activity at the merchant’s site or at PayPal’s site and information stored in cookies. When an unusual pattern is revealed such as a string of purchases or a sudden change in geographic location, the activity gets turned into a “feature,” or a rule that can be applied in real time to stop purchases that fit suspicious activity. This “feature” can then be applied to stop purchases that meet its criteria. Since PayPal contains approximately over 1.1 PT of customer data, deep learning algorithms are able to analyze potentially tens of thousands of latent features that can help curb fraudulent activity.
The cumulative effect of using deep learning is largely beneficial at preventing further losses. According to a LexisNexus report, PayPal’s fraud rate is at a remarkably low 0.32% against the financial industry standard of 1.32%.