Graph-Based Account Anomaly Detection for Off-Network Payments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to detect fraudulent transactions at the account level, allowing fraudsters to exploit the system, and conventional fraud detection systems are inefficient in detecting fraud at the transaction level, failing to identify and prevent fraudulent transactions at the account level.
Innovation Solution
Implementing a graphical modeling system that utilizes machine learning algorithms to detect fraudulent transactions at the account level by analyzing the account-to-account ecosystem and identifying fraudulent transactions by analyzing the account level.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional fraud detection systems evaluate transactions at the transaction level, then individual fraudulent transactions can be identified, but bad actors cannot be detected and repeated fraud attempts are not prevented
Solution Approach 1:
The patent transitions from transaction-level detection to account-level detection by building a graphical model that represents accounts as nodes and transactions as edges. This dimensional shift allows the system to evaluate patterns across multiple transactions associated with a single account, enabling detection of bad actors through aggregate behavior analysis rather than isolated transaction assessment.
Solution Approach 2:
The system merges multiple transaction data points into a unified graphical model where all transactions associated with an account are combined and analyzed together. By aggregating transaction features, decline patterns, and account behavior into a single graphical representation, the system achieves comprehensive fraud detection at the account level while maintaining manageable complexity through structured data integration.
2Loss of time
If fraud detection is performed only after fraud is committed at the transaction level, then individual fraud cases can be identified, but early-stage fraudulent behavior cannot be prevented
Solution Approach 1:
The system performs preliminary fraud detection by analyzing account-level patterns before fraudulent transactions are fully executed. The graphical model continuously evaluates account behavior, decline patterns, and transaction features in real-time, enabling early identification of suspicious accounts. This preliminary detection allows the system to flag potential fraudsters before they can complete multiple fraudulent transactions, reducing both time loss and improving reliability.
3Loss of information
If transaction data is analyzed without graphical modeling, then processing is simpler, but account-level fraud patterns cannot be identified
Solution Approach 1:
The graphical model segments the complex account-level fraud detection problem into manageable components: accounts are represented as nodes, transactions as edges, and various features (decline patterns, transaction amounts, frequencies) are segmented into separate analytical layers. This segmentation allows the system to process and analyze each feature type independently while maintaining the overall account-level context, thereby preserving comprehensive fraud pattern information without overwhelming system complexity.
Data Source
AI summary
A graphical modeling computer system is configured to: (i) receive transaction data for a plurality of transactions including declined transactions and approved transactions; (ii) parse the transaction data to identify each pair of a respective payment node and a respective funding node associated with each transaction of the plurality of transactions; (iii) build a graphical model showing all transactions of the plurality of transactions initiated from each funding node to a corresponding payment node; (iv) train the graphical model by labeling each transaction as either a completed transaction or a declined transaction; (v) train the graphical model by applying fraud labels to each transaction, by mapping aggregated parameters to each payment node, and by associating transaction features to each incoming or outgoing transaction for the payment nodes; and (vi) output from the trained graphical model an indication of the payment nodes associated with suspicious activity.