Event Stream Joining for Real-Time Fraud Feature Vectors

Solution Overview

Problem

Conventional methods for generating input data for advanced machine-learning models, particularly for real-time fraud detection, are constrained by memory and bandwidth limitations, requiring complex and time-consuming feature retrieval processes.

Innovation Solution

A system and method for generating input feature vectors by joining events from multiple event streams, allowing for efficient feature vector generation without relying on costly upstream event joining operations, and enabling automatic fraud detection across multiple systems.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional feature retrieval processes are used for real-time fraud detection, then memory access requirements and bandwidth consumption increase, but feature generation accuracy is maintained

Engineering Contradiction:

Improvefeature generation speedVSAvoidmemory resources

Core Design Contradiction:

ProductivityVSQuantity of substance

Solution Approach 1:

The patent segments the feature generation process into two distinct components: an offline feature generation system that pre-processes and stores feature data in archives, and an online feature retrieval system that quickly accesses pre-generated features. This segmentation allows the online system to avoid computationally intensive processing while maintaining real-time performance, thereby reducing memory access requirements and bandwidth consumption during fraud detection operations.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent implements preliminary action by pre-generating and storing feature data in offline archives before real-time fraud detection is needed. The offline feature generation system processes historical data and creates feature vectors that are stored for rapid retrieval. This preliminary processing eliminates the need for complex real-time feature computation during online fraud detection, significantly reducing memory bandwidth requirements while maintaining feature generation accuracy.

Inventive Principle:

Principle #10Preliminary action

2Loss of time

If offline feature generation techniques are used, then feature retrieval time increases, but computational cost during real-time processing is reduced

Engineering Contradiction:

Improvefeature retrieval timeVSAvoidfeature retrieval process

Core Design Contradiction:

Loss of timeVSDevice complexity

Solution Approach 1:

The patent implements preliminary action by pre-generating and storing feature data in offline archives before real-time fraud detection is needed. The offline feature generation system processes historical data and creates feature vectors that are stored for rapid retrieval. This preliminary processing eliminates the need for complex real-time feature computation during online fraud detection, significantly reducing memory bandwidth requirements while maintaining feature generation accuracy.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary component: the offline feature archive that acts as a buffer between raw event data and the online fraud detection system. This intermediary stores pre-processed feature data that can be quickly accessed during real-time operations, decoupling the computational complexity from the time-critical path while maintaining fast feature retrieval.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Measurement precision

If complex feature retrieval processes are implemented, then feature accuracy is maintained, but system bandwidth consumption increases

Engineering Contradiction:

Improvefeature accuracyVSAvoidbandwidth consumption

Core Design Contradiction:

Measurement precisionVSLoss of energy

Solution Approach 1:

The patent segments the feature generation process into two distinct components: an offline feature generation system that pre-processes and stores feature data in archives, and an online feature retrieval system that quickly accesses pre-generated features. This segmentation allows the online system to avoid computationally intensive processing while maintaining real-time performance, thereby reducing memory access requirements and bandwidth consumption during fraud detection operations.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent creates copies of feature data by generating and storing feature vectors in offline archives before they are needed for real-time fraud detection. These pre-generated feature copies are stored in an optimized format that allows rapid retrieval without requiring complex processing during online operations, thereby maintaining feature accuracy while reducing bandwidth consumption during real-time operations.

Inventive Principle:

Principle #26Copying

Data Source

PatentUS20250356375A1Systems and methods for using feature computation systems to join event streams

Publication Date: 2025.11.20 STRIPE LLC

AI summary

One method includes detecting a condition associated with an event stream of a first plurality of event streams associated with a first system; determining that the event stream is associated with a second system; identifying a second plurality of event streams associated with the second system; generating, based on the condition, an event data structure comprising a first set of events identified from the first plurality of event streams and a second set of events identified from the second plurality of event streams; converting the event data structure into at least two feature vectors corresponding to the first system and the second system for one or more machine-learning models; and executing the one or more machine-learning models using the at least two feature vectors as input and outputting a likelihood of fraud for the first system or the second system.