Fuzzy Peer Grouping for Real-Time Mule Account Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional clustering methods oversimplify the grouping of individuals or transactions, potentially missing nuanced relationships and multi-layered structures in anti-money laundering (AML) investigations, leading to incomplete and low-accuracy peer grouping.
Innovation Solution
A smart peer grouping system that allows entities to belong to multiple clusters and employs soft/fuzzy clustering with Gaussian mixture models and the Expectation Maximization (EM) algorithm to capture complex patterns and multi-layered structures in transaction data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional clustering methods are used to group entities, then the system is simple and easy to implement, but the accuracy and completeness of peer grouping deteriorates due to oversimplification of complex relationships
Solution Approach 1:
The patent transforms hard clustering parameters into soft/fuzzy parameters, allowing entities to have probabilistic memberships across multiple clusters rather than binary assignments. This parameter transformation enables the system to capture nuanced relationships and multi-layered structures in transaction data, directly improving peer grouping accuracy while managing complexity through mathematical formalization
Solution Approach 2:
The patent implements dynamic cluster assignments where entities can belong to multiple clusters simultaneously with varying degrees of membership. This dynamic approach allows the clustering system to adapt to complex patterns in AML data, improving detection accuracy by representing entities' multifaceted nature rather than forcing static single-cluster assignments
2Loss of information
If traditional hard clustering is used where entities belong to only one cluster, then the computational process is simple, but the system fails to capture nuanced relationships and multi-layered structures in transaction data
Solution Approach 1:
The patent changes the membership parameter from binary (0 or 1) to continuous probability values between 0 and 1, allowing entities to partially belong to multiple clusters. This parameter change preserves nuanced relationship information by quantifying the degree of association between entities and clusters, preventing information loss that occurs in hard clustering
Solution Approach 2:
The patent creates a composite clustering approach that combines Gaussian mixture models with Expectation Maximization algorithm. This composite methodology integrates probabilistic modeling with iterative optimization, enabling the system to capture complex patterns and multi-layered structures while providing a unified framework for soft clustering in AML investigations
3Reliability
If soft/fuzzy clustering with Gaussian mixture models and EM algorithm is used, then the detection accuracy of sophisticated money laundering techniques is improved, but the computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary clustering using Gaussian mixture models to establish probabilistic associations before detailed AML analysis. This preliminary action pre-computes the soft cluster assignments and probability distributions, reducing the computational burden during real-time transaction monitoring while maintaining high detection reliability for sophisticated money laundering techniques
Data Source
AI summary
A system is adapted to automatically identify suspected mule accounts. It includes a processor performing operations: identifying a number of desired clusters for grouping entities; for each dimension in a multidimensional space, defining each cluster as a Gaussian distribution in each dimension. For each entity: for each cluster: calculating a distance between the entity and the cluster, and a probability that the entity belongs to the cluster; and recalculating the Gaussian distributions until each entity belongs to at least one. The operations also include, for an entity, in real time: receiving a transaction associated with the entity; based on a cluster to which the entity belongs, determining a peer anomaly score indicative of a probability that the transaction is anomalous; and if the peer anomaly score exceeds a threshold value, reporting the transaction and the entity to a user.