Fraud Risk Evaluation System Using Sigmoidal Score Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current credit card verification systems in electronic commerce transactions face challenges in accurately assessing fraud risk, particularly in online transactions, where address and identity verification are insufficient, leading to increased risk of fraudulent activities and inefficient manual screening methods that impact business operations.
Innovation Solution
A system that evaluates fraud risk by applying transaction information to multiple mathematical models, transforming raw scores using sigmoidal functions, and combining risk estimates through fusion proportions to produce a single optimized risk estimate, dynamically adjusting to changes in transaction patterns and fraud behaviors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional address verification systems (AVS) are used for credit card transactions, then basic address matching can be achieved, but fraud detection accuracy is insufficient leading to high false positive and false negative rates
Solution Approach 1:
The fraud detection system segments the verification process into multiple independent mathematical models, each evaluating specific risk factors separately. These models include but are not limited to: velocity checks (transaction frequency analysis), geographic consistency checks, purchase pattern analysis, and device fingerprinting. Each model produces an independent risk score that is then combined, allowing the system to achieve high detection accuracy while maintaining reliability through diversified evaluation dimensions.
Solution Approach 2:
The system employs a composite evaluation framework that integrates multiple mathematical models with different strengths. Each model acts as a component with specific detection capabilities, and their combined output creates a comprehensive fraud assessment. This composite approach combines the advantages of individual models while compensating for their individual weaknesses, achieving both high accuracy and reliability in fraud detection.
2Reliability
If manual screening methods are implemented to reduce fraud risk, then fraud detection capability improves, but operational overhead and processing time increase significantly
Solution Approach 1:
The system implements self-service fraud detection through automated mathematical models that independently evaluate transactions without human intervention. The models automatically analyze transaction data, apply risk assessment algorithms, and generate fraud probability scores. This automation maintains high fraud detection capability while eliminating the operational overhead and time delays associated with manual screening, thereby preserving transaction processing efficiency.
Solution Approach 2:
The system replaces manual mechanical screening processes with automated computational mathematical models. Instead of human analysts reviewing transactions, the system uses algorithms including neural networks, decision trees, and statistical analysis to automatically assess fraud risk. This substitution maintains or improves detection capability while dramatically increasing processing speed and reducing operational costs.
3Measurement precision
If multiple verification checks are applied to each transaction, then fraud detection accuracy improves, but system complexity and processing time increase
Solution Approach 1:
The system implements dynamic verification where the application of mathematical models is adaptive rather than static. The system adjusts which models are applied and their weighting based on transaction characteristics, risk indicators, and historical data. For low-risk transactions, fewer models are activated, while high-risk transactions trigger comprehensive multi-model analysis. This dynamic approach maintains high detection accuracy while reducing average system complexity and processing time.
Solution Approach 2:
The system changes verification parameters dynamically based on transaction context. Instead of applying all mathematical models uniformly to every transaction, the system adjusts model activation, threshold settings, and evaluation depth based on initial risk indicators. This parameter adaptation allows the system to maintain high accuracy for complex cases while simplifying processing for routine transactions, thereby reducing overall system complexity.
Data Source
AI summary
According to one aspect, transaction information is received and applied to multiple fraud risk mathematical models that each produce a respective raw score, which are transformed with respective sigmoidal transform functions to produce optimized likelihood of fraud risk estimates to provide to a merchant. In one embodiment, the respective fraud risk estimates are combined using fusion proportions that are associated with the respective risk estimates, producing a single point risk estimate, which is transformed with a sigmoidal function to produce an optimized single point risk estimate for the transaction. The sigmoidal functions are derived to approximate a relationship between risk estimates produced by fraud risk detection models and a percentage of transactions associated with respective risk estimates, where the relationship is represented in terms of real-world distributions of fraudulent transaction and non-fraudulent transaction. One embodiment is directed to computing respective risk test penalties for multiple risk tests in one or more of the multiple fraud risk mathematical models used to estimate the likelihood of fraud, given a certain pattern of events represented by the transaction information, wherein the respective risk test penalties are computed as the inverse of the sum of one and a false positive ratio for the respective risk test.