Fraud Detection System Using Multi-Model Risk Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional fraud detection methods for online transactions, particularly CNP transactions, are inadequate as they rely on physical inspection and are ineffective in identifying fraudulent activities without the cardholder's presence, leading to increased fraudulent transactions.
Innovation Solution
A system that assesses online transactions by analyzing various data values characterizing the identities of both parties involved, including IP addresses, payment mechanisms, and personalized information, using multiple fraud detection models to generate a risk score and authorize or reject transactions based on threshold scores, with the option for manual review.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional fraud detection techniques requiring physical manipulation of the card are used, then card security is maintained in physical transactions, but fraudulent CNP transactions cannot be detected
Solution Approach 1:
The patent replaces physical card manipulation mechanisms with data analysis mechanisms. Instead of requiring physical inspection of the card, the system analyzes multiple data values including IP addresses, device characteristics, transaction patterns, and message content to detect fraud in CNP transactions where the cardholder is not physically present.
Solution Approach 2:
The patent introduces an intermediary fraud detection system that acts as a mediator between the transaction parties and the authorization system. This intermediary analyzes multiple data sources and generates a fraud score to determine whether to proceed with the transaction, bridging the gap between remote CNP transactions and traditional fraud detection.
2Reliability
If multiple data values and fraud detection models are analyzed for each transaction, then fraud detection accuracy is improved, but processing time and system complexity increase
Solution Approach 1:
The patent segments the fraud detection process into multiple independent components: data collection module, data analysis module with multiple fraud detection models, scoring module, and decision module. Each component performs a specific function, allowing the system to analyze multiple data values thoroughly while maintaining modularity and manageability.
Solution Approach 2:
The patent creates a universal fraud detection system that handles multiple types of data values (IP addresses, device information, transaction patterns, message content) through a single multi-functional platform. The system can process various data types using different fraud detection models, providing comprehensive fraud detection without requiring separate systems for each data type.
3Reliability
If comprehensive data analysis is performed on all transactions, then fraudulent transactions are detected early, but legitimate transactions may be incorrectly flagged
Solution Approach 1:
The patent uses parameter changes by adjusting the fraud score thresholds and weights assigned to different data values based on the specific transaction context. The system dynamically modifies analysis parameters to balance detection sensitivity and false positive rates, allowing comprehensive analysis while reducing incorrect flags for legitimate transactions.
Solution Approach 2:
The patent implements feedback mechanisms where the results of fraud detection are used to refine future analysis. The system learns from detected fraud patterns and adjusts its analysis parameters, improving accuracy over time while reducing false positives through continuous optimization based on actual transaction outcomes.
Data Source
AI summary
The subject matter disclosed herein provides methods for detecting and mitigating fraudulent transactions conducted over a network. This method can access one or more data values associated with an online transaction between a first party and a second party. The first party can initiate the online transaction. The data values can characterize an identity of the first party and the second party. A score representing a likelihood that the online transaction is fraudulent can be generated. The generation of the score can be based on one or more models for detecting fraud using the data values. The score can be compared to one or more threshold scores. The online transaction can be authorized or rejected based on the comparing. Related apparatus, systems, techniques, and articles are also described.