Dimensional Reduction for Check Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for detecting fraudulent checks face challenges due to high storage and computing requirements, leading to slow processing and increased costs, as they need to store and retrieve large volumes of check serial number histories to identify out-of-sequence features.
Innovation Solution
A dimensional reduction integrated fraud management system that maps check serial numbers into a memory map of digit usage over time, using an array of position-based aggregators to store digit usage instead of actual check numbers, allowing for faster and more efficient fraud detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all check serial numbers are stored in a database to identify out-of-sequence features, then fraud detection accuracy is improved, but storage requirements and computing resources increase significantly
Solution Approach 1:
The patent extracts only the essential features from complete serial number storage - specifically tracking the range of serial numbers (minimum and maximum values) and detecting out-of-sequence features. This extraction approach maintains fraud detection capability while dramatically reducing storage requirements from storing millions of individual serial numbers to storing only range boundaries.
Solution Approach 2:
The patent changes the parameters being tracked from complete serial number values to aggregated range parameters (min, max) and out-of-sequence indicators. This parameter transformation allows the system to maintain detection accuracy for fraudulent patterns while reducing the data volume to a manageable size.
2Reliability
If complete check serial number history is retrieved in real time, then fraud detection capability is improved, but processing speed decreases
Solution Approach 1:
The patent performs preliminary aggregation of serial number ranges and out-of-sequence features during data ingestion, before real-time fraud detection is needed. This pre-processing creates ready-to-use range boundaries and anomaly indicators that can be quickly queried during transaction processing, eliminating the need for real-time historical data retrieval.
Solution Approach 2:
Instead of querying historical serial number data during fraud detection, the system inverts the approach by pre-computing and storing the essential detection features (ranges and out-of-sequence markers). This inversion transforms the system from a query-heavy architecture to a storage-efficient architecture where detection logic operates on pre-processed data.
3Reliability
If all check numbers are stored and processed, then comprehensive fraud analysis is achieved, but computing requirements and costs increase
Solution Approach 1:
The patent extracts only the computationally essential features from complete serial number data - specifically the range boundaries and out-of-sequence anomalies. This extraction eliminates the need for computationally intensive operations on millions of individual serial numbers while preserving the ability to detect fraudulent patterns.
Solution Approach 2:
The patent applies partial action by focusing computational resources on tracking only the critical aspects of serial number sequences (ranges and outliers) rather than processing complete historical data. This partial approach achieves sufficient fraud detection capability with dramatically reduced computing requirements.
Data Source
AI summary
A system includes collecting serial numbers over a period of time, and constructing a matrix indicating which digits have been used at which positions. For a second period of time, the system collects a second group of serial numbers and, for each of these serial numbers, identifies a feature of the serial number by comparing it against the matrix, and automatically updates the matrix with the digits and digit positions of the serial number. The features are received into an artificial intelligence model as training data. Further, the system collects a third group of serial numbers and, for each serial number of the third group, identifies a feature of the serial number by comparing it against the matrix. These features are then received into the artificial intelligence model, which determines a risk score. The matrix is then updated with the digits and digit positions of the serial number.