The uncertain case of credit card fraud detection

Abstract

Uncertainty is inherent in many real-time event-driven applications. Credit card fraud detection is a typical uncertain domain, where potential fraud incidents must be detected in real time and tagged before the transaction has been accepted or denied. We present extensions to the IBM Proactive Technology Online (PROTON) open source tool to cope with uncertainty. The inclusion of uncertainty aspects impacts all levels of the architecture and logic of an event processing engine. The extensions implemented in PROTON include the addition of new built-in attributes and functions, support for new types of operands, and support for event processing patterns to cope with all these. The new capabilities were implemented as building blocks and basic primitives in the complex event processing programmatic language. This enables implementation of eventdriven applications possessing uncertainty aspects from different domains in a generic manner. A first application was devised in the domain of credit card fraud detection. Our preliminary results are encouraging, showing potential benefits that stem from incorporating uncertainty aspects to the domain of credit card fraud detection.

Uncertainty-Aware Event Analytics over Distributed Settings

“…On the other hand, should one simply use the extension of Proton provided by Correia et al [7], she will need to accumulate all data to the query site before being able to check whether the uncertainty criterion applies and whether events can be pruned because of it.…”

“…The first three columns of Table 1 include the name of the distribution, its Probability Density Function (PDF) and an explanation of its parameters, respectively. The last column of Table 1 refers to our proof-of-concept (Section 5), highlighting which of the cited distributions are present in the uncertain extension of IBM Proton [7]. We believe that as uncertainty-aware CEP gets more and more attention, such functionality will be provided by other engines as well.…”

“…• As a proof-of-concept of the applicability of our techniques in real systems, we show how we bridge the gap between two recent prototypes. The FERARI streaming multi-cloud platform [12,13], which provides communication-efficient event analytics over distributed settings but neglects uncertainty aspects, and the work by Correia et al [7] which elaborates on handling event uncertainty but requires continuous event transmissions, not accounting for efficiency over distributed settings. Both prototypes utilize IBM PROactive Technology ONline (Proton) [1] as their CEP engine and each extend it towards the respective aspect.…”

The Event Crowd

“…For example, in environmental monitoring, the data is processed from various sensors to understand pollution levels so that immediate action can be taken to counter its negative effects. Similarly, other fields like financial [2], transportation [3], and manufacturing require the processing of data in near-real-time. Event-based systems process data from different sources like sensors and social media feeds where the event data can be veracious and have inherent uncertainties.…”

“…These uncertainties can be due to multiple reasons including incomplete event streams, erroneous event recognition and imprecise event patterns [4]. Ivo et al [2] distinguish three types of uncertainties in event data i.e. uncertainty in event content, occurrence, and rules.…”