Ratio Striping Engine for Dynamic Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fraud detection systems in payment card networks struggle to detect low-level fraud attacks and changing tactics, as they rely on static models that do not adapt to evolving fraudulent activities, leading to undetected network-based fraud events, increased network load, and computational burdens.
Innovation Solution
A computing system and method that employs a ratio striping engine to process scored payment card transaction authorization requests, generating cumulative metrics and ratio striping values over time, which are used to update parameters of a downstream fraud detection model, enabling real-time adaptation and improved fraud detection through machine learning algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If static fraud detection models are used, then device complexity is reduced, but adaptability to changing fraud tactics deteriorates
Solution Approach 1:
The patent implements dynamic fraud detection by continuously updating model parameters based on incoming transaction data. The system transitions from static models to dynamic models that adapt in real-time, using rolling time windows and cumulative metrics to capture evolving fraud patterns while maintaining manageable complexity through structured data organization.
Solution Approach 2:
The system changes model parameters dynamically by updating cumulative metrics and ratio striping values based on incoming transactions. These parameter changes enable the model to adapt to new fraud tactics without requiring complete model retraining, thus improving adaptability while controlling complexity through incremental updates.
2Measurement precision
If comprehensive fraud analysis is performed on all transactions, then detection precision is improved, but processing time increases
Solution Approach 1:
The patent segments transactions into different fraud score ranges (stripes) and processes them differently. Low-risk transactions receive minimal processing while high-risk transactions undergo more comprehensive analysis. This segmentation maintains high detection precision for fraudulent transactions while reducing average processing time through selective detailed analysis.
Solution Approach 2:
The system applies partial analysis to low-risk transactions and excessive (comprehensive) analysis to high-risk transactions. By using ratio striping values to identify transactions requiring full scrutiny, the system achieves high detection precision where needed while minimizing processing time for the majority of legitimate transactions through streamlined processing.
3Productivity
If real-time fraud detection is implemented, then productivity is improved, but computational burden increases
Solution Approach 1:
The system performs preliminary fraud scoring and ratio striping calculations on incoming transactions before applying the full fraud detection model. This preliminary action filters out obviously legitimate transactions, reducing the computational burden on the main detection system while maintaining real-time detection capability for suspicious transactions.
Solution Approach 2:
The patent introduces ratio striping values as an intermediary metric between raw transaction data and final fraud detection. These intermediary calculations organize data into meaningful patterns that reduce the computational complexity of the final detection step, enabling real-time processing with reduced computational burden.
Data Source
AI summary
A method and system for detecting fraudulent network events in a payment card network are provided. A plurality of scored payment card transaction authorization requests are received, originating from a plurality of merchants, and at least one data structure is generated. The data structure sorts the scored authorization requests into fraud score range stripes. The data structure is parsed over a plurality of time periods, and at least one cumulative metric is calculated for each of the time periods in each fraud score range stripe. A plurality of ratio striping values is determined, and a set of feature inputs is generated using the ratio striping values. A second fraud detection model is applied to the scored authorization requests. Parameters of the second fraud detection model are configured to change based on the generated set of feature inputs.