Transaction Clustering for Scalable Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fraud detection systems are not scalable to secure consumer applications effectively, as they are designed for lower volume deployments and are not cost-effective for millions of users, leading to inadequate protection against identity theft and online fraud.
Innovation Solution
A method and system for determining a clustering metric by analyzing transaction similarities through intersection and union sets, allowing for accurate clustering and anomaly detection in online transactions, which can be applied to various computer networks and communications systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If enterprise authentication solutions are deployed to secure consumer applications, then security and fraud prevention are improved, but cost and scalability deteriorate due to the high volume of transactions
Solution Approach 1:
The patent segments the fraud detection process into multiple components: (1) clustering transactions into groups based on similarity, (2) computing clustering metrics for each group, and (3) analyzing anomalies within clusters. This segmentation allows the system to handle millions of transactions by processing them in organized groups rather than individually, improving scalability while maintaining security.
Solution Approach 2:
The patent changes the parameter of transaction analysis from individual transaction evaluation to cluster-based evaluation. By computing clustering metrics (such as density, diameter, and separation) and analyzing transactions within clustered groups, the system reduces the computational burden while maintaining fraud detection accuracy, thus resolving the scalability contradiction.
2Reliability
If traditional fraud detection methods are used, then security measures are implemented, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent performs preliminary clustering of transactions before detailed fraud analysis. By pre-grouping transactions into clusters based on similarity metrics, the system reduces the complexity of subsequent fraud detection operations. This preliminary organization allows security analysis to focus on relevant transaction groups rather than all transactions individually, reducing computational complexity while maintaining security.
3Measurement precision
If clustering metrics are computed for all transactions, then accurate fraud detection is achieved, but processing time and computational resources increase
Solution Approach 1:
The patent computes clustering metrics partially by focusing on representative transactions within each cluster rather than all transactions. The system identifies key transactions that characterize each cluster and computes metrics based on these representatives, achieving sufficient clustering accuracy without the excessive processing time required for complete transaction-by-transaction metric computation.
Data Source
AI summary
A method of determining a clustering metric includes receiving a first set of transactions and a second set of transactions. For transaction i of the first set and transaction j of the second set, the method includes (a) determining an intersection set, (b) determining a union set; (c) computing a common linkage between transaction i and transaction j equal to the intersection set divided by the union set, and (d) incrementing index j and repeating steps (a)-(c). The method also includes (e) summing the common linkages between transaction i and the transactions of the second set, (f) normalizing the sum of the common linkages by a number of the second set, and (g) incrementing index i and repeating steps (a)-(f). The method further includes (h) summing the normalized common linkages and (i) normalizing the sum of the normalized common linkages by a number of the first set.