Co-occurrence Consistency Analysis for Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for detecting fraudulent transactions face challenges in scalable variable selection, as they often rely on greedy methods that can result in local maxima and assume a single reason for fraud, failing to account for multiple variable groups that may indicate different forms of fraud.
Innovation Solution
The proposed solution involves a co-occurrence consistency analysis method that uses a learning component to extract relevant variables by quantifying co-operative strength between pairs and forming cliques, allowing for the identification of multiple variable groups that can predict fraud, thereby avoiding the limitations of greedy methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If exhaustive search for variable groups is performed to achieve best model prediction accuracy, then prediction accuracy is improved, but computing time and resource requirements increase significantly
Solution Approach 1:
The patent segments the variable selection process into multiple independent cliques, where each clique represents a distinct group of variables that work together effectively. Instead of evaluating all possible variable combinations exhaustively, the method divides the search space into manageable segments (cliques) that can be evaluated independently and in parallel, significantly reducing computing time while maintaining prediction accuracy.
Solution Approach 2:
The patent performs preliminary actions by pre-identifying and storing multiple cliques of variables before model training. These cliques are pre-computed based on variable correlations and co-occurrence patterns in the training data, allowing the system to quickly select appropriate variable groups during deployment without performing exhaustive searches at runtime.
2Productivity
If greedy methods are used for variable selection to reduce computing time, then computing efficiency is improved, but the method converges to local maxima and may miss optimal variable groups
Solution Approach 1:
The patent introduces dynamics by iteratively refining cliques through multiple passes of evaluation and adjustment. Rather than using a static greedy approach, the method dynamically adjusts variable groupings by evaluating different clique configurations, merging overlapping cliques, and splitting inadequate ones, allowing the system to escape local maxima and converge to better solutions.
Solution Approach 2:
The patent implements feedback mechanisms where the performance of each clique is evaluated based on model prediction accuracy, and this feedback is used to guide subsequent clique formation and refinement. Cliques that perform poorly are modified or discarded, while successful patterns are reinforced and expanded, creating a feedback-driven optimization process that improves both efficiency and accuracy.
3Device complexity
If a single variable group is selected for fraud detection, then model simplicity is improved, but the ability to detect different forms of fraud is reduced
Solution Approach 1:
The patent applies universality by designing a multi-clique architecture where each clique serves as an independent fraud detection specialist for different fraud patterns. The system can selectively activate specific cliques based on the transaction characteristics being analyzed, allowing a single system to handle multiple fraud types (telecommunications fraud, credit card fraud, ATM fraud) without requiring separate complex models for each.
Solution Approach 2:
The patent segments the fraud detection capability into multiple specialized cliques, where each clique captures a specific pattern or type of fraudulent behavior. This segmentation allows the system to maintain simplicity within each individual clique while achieving high versatility through the collective power of multiple cliques, each optimized for detecting particular fraud forms.
Data Source
AI summary
A method of modeling includes quantifying a co-operative strength value for a plurality of pairs of variables, and identifying a clique of at least three variables based on a graph of the co-operative strength values of a plurality of pairs of variables. The method also includes selecting a first pair of variables of the plurality of pairs of variables having a high co-operative strength value. A second clique may also be identified. A model of the first clique and a model of the second clique are made. The outputs of these models are combined to form a combined model which is used to make various decisions with respect to real time data.