Account Activity Fingerprinting for Rapid Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to adequately detect account impersonation and fraudulent activity in a timely manner, leading to prolonged unauthorized use and significant fraudulent transactions, particularly in industries like healthcare where impersonation can result in weeks of improper actions before detection.
Innovation Solution
A method involving querying a database to generate tables of historical transaction patterns over different time periods, calculating similarity measures, and identifying accounts with significant deviations as fraudulent candidates, enabling real-time or near real-time detection and prevention of further transactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If traditional account monitoring methods are used, then system complexity remains low, but detection time for fraudulent activity becomes excessively long (weeks)
Solution Approach 1:
The system pre-calculates and stores expected transaction patterns for each account based on historical data before fraud occurs. When a transaction occurs, the system immediately compares it against the pre-established pattern, enabling rapid detection without requiring complex real-time analysis during the fraud event itself.
Solution Approach 2:
The monitoring system uses each account's own historical transaction data to generate its expected pattern, allowing the account's past behavior to serve as its own baseline for detection. This self-referential approach eliminates the need for external complex analysis models while enabling fast, accurate fraud detection.
2Reliability
If comprehensive transaction monitoring is implemented, then fraud detection capability improves, but computing resources and processing time increase significantly
Solution Approach 1:
The system extracts only the essential pattern information from historical transactions (such as typical transaction amounts, frequencies, and timing) and stores this condensed pattern data rather than the complete transaction history. This extraction reduces the amount of data that needs to be processed during fraud detection while maintaining detection accuracy.
Solution Approach 2:
The system transforms detailed transaction data into simplified pattern parameters (expected amount ranges, frequency intervals, temporal patterns) that are much easier and faster to compare against new transactions. This parameter transformation enables comprehensive monitoring with reduced computational overhead.
3Productivity
If real-time fraud detection is implemented, then fraudulent transactions are prevented, but system response time and processing speed requirements increase
Solution Approach 1:
By pre-computing expected transaction patterns from historical data and storing them in a readily accessible format, the system eliminates the need for slow real-time pattern learning during transaction processing. The pattern generation happens beforehand, allowing instantaneous comparison during actual transactions.
Solution Approach 2:
The system creates a simplified copy or representation of the account's normal transaction pattern (the expected pattern table) that can be rapidly compared against new transactions without re-analyzing the complete historical data each time. This pattern copy enables fast detection while maintaining accuracy.
Data Source
AI summary
Methods, systems, apparatuses, and computer program products for discovering fraudulent account usage are disclosed. In an example, a database is queried to identify a list of accounts and transaction categories. A first table is generated based on the transaction categories. A second table is generated by populating the first table based on a response to querying the database for a first set of historical transactions over a first period of time. A third table is generated by populating the first table based on a response to querying the database for a second set of historical transactions over a second period of time. A measure of similarity for each of the accounts is generated based on the second table and the third table. For each account with a corresponding measure of similarity below a threshold, the account is identified as a fraudulent account candidate.