Dynamic Account Fraud Detection via Adaptive Threshold Rules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches to identifying illegitimate accounts in online services are often ineffective as they rely on static rules that can be easily evaded by fraudulent users, leading to compromised financial instruments and reduced user experience.
Innovation Solution
A dynamic identification system that analyzes features and feature combinations of accounts within a specified time frame, determines feature metrics, and implements rules based on threshold values to identify illegitimate accounts, using a combination of machine learning and manual effort to adjust thresholds for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If static rules are used to identify illegitimate accounts, then the system is simple to implement, but the identification accuracy deteriorates as fraudulent users can easily evade the rules
Solution Approach 1:
The patent implements dynamic rule adjustment by continuously monitoring account features (such as login locations, device types, transaction patterns) and automatically updating identification rules based on detected fraudulent behaviors. The system transitions from static rules to dynamic rules that adapt to evolving fraud tactics, thereby improving identification accuracy without requiring manual rule rewriting for each fraud scheme.
Solution Approach 2:
The system incorporates feedback mechanisms where identified fraudulent accounts and their behaviors are fed back into the rule generation process. Manual review of flagged accounts and analysis of successful fraud evasion attempts provide feedback that refines future rule sets, creating a continuous improvement cycle that enhances reliability over time.
2Reliability
If manual review of all accounts is performed, then identification accuracy is high, but processing time and operational cost increase significantly
Solution Approach 1:
The patent segments the account review process into three tiers: automated rule-based screening for high-confidence fraud cases, manual review for ambiguous or high-risk cases, and monitoring for potential future fraud. This segmentation allows the system to maintain high accuracy for critical cases while rapidly processing the majority of accounts through automation, significantly reducing overall processing time.
Solution Approach 2:
The system applies partial manual review by selectively reviewing only those accounts that meet specific risk criteria or have been flagged by automated detection, rather than manually reviewing all accounts. This partial action approach maintains high identification accuracy for problematic accounts while avoiding the time-consuming process of manual review for all accounts.
3Measurement precision
If multiple feature metrics are analyzed for each account, then identification precision improves, but computational resources and processing complexity increase
Solution Approach 1:
The patent performs preliminary actions by pre-calculating and storing feature metrics for accounts during their normal operation. Common features such as login frequency, geographic patterns, device fingerprints, and transaction histories are continuously monitored and stored in a structured format. When fraud detection is triggered, these pre-computed metrics are immediately available for analysis, eliminating the need for real-time calculation of all features and reducing processing complexity.
Solution Approach 2:
The system applies local quality by analyzing different feature metrics with different levels of depth depending on the specific account context and risk level. High-risk accounts receive comprehensive multi-feature analysis, while low-risk accounts undergo simplified checking. This localized approach to quality control maintains measurement precision where needed while reducing overall processing complexity.
Data Source
AI summary
Systems, methods, and non-transitory computer-readable media can receive a set of accounts associated with a specified time frame. One or more features and one or more feature combinations can be analyzed for each account in the set. Feature metrics for the one or more features and the one or more feature combinations can be determined for each account in the set. Threshold values for the feature metrics can be acquired. At least one rule can be implemented based on at least some of the feature metrics and at least some of the threshold values.