Electronic Service Filter Optimization via Dynamic Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems face challenges in efficiently identifying and filtering out undesired electronic transactions, such as chargeback fraud, which lead to unnecessary computational resources and processing delays due to the difficulty in human analysis and the high frequency of false positives.
Innovation Solution
Implementing a machine learning tree-based algorithm that analyzes transaction history to determine optimal cutoff values for risk scores, calculating precision scores, and recommending threshold adjustments for processing rules to improve performance metrics like reducing computational resources and false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual analysis and traditional filtering methods are used to identify undesired transactions, then human review can detect complex fraud patterns, but computational resources are wasted and processing delays increase due to high false positive rates
Solution Approach 1:
The patent segments the fraud detection process into multiple hierarchical levels: (1) automated machine learning models perform initial filtering of high-risk transactions, (2) rule-based systems handle intermediate-risk cases, and (3) human analysts focus only on complex cases that escape automated systems. This segmentation reduces the volume of transactions requiring human review while maintaining detection accuracy.
Solution Approach 2:
The patent introduces machine learning models and automated rule engines as intermediary systems between raw transaction data and human analysts. These intermediaries pre-process and filter transactions, providing human reviewers with prioritized, pre-analyzed case lists that reduce false positives and improve processing efficiency.
2Object-affected harmful factors
If traditional filtering rules with fixed thresholds are applied to block high-risk transactions, then chargeback fraud can be reduced, but false positives increase leading to legitimate transactions being blocked
Solution Approach 1:
The patent dynamically adjusts filtering thresholds and risk score weights based on transaction context, historical data, and real-time patterns. Instead of fixed thresholds, the system modifies parameters such as risk score cutoffs, feature importance weights, and rule sensitivity levels to optimize the balance between fraud prevention and false positive reduction for different transaction types and time periods.
Solution Approach 2:
The patent implements dynamic, adaptive filtering rules that automatically adjust their sensitivity and thresholds based on learned patterns from transaction data. The system continuously updates risk models and rule parameters to adapt to evolving fraud patterns while maintaining appropriate false positive rates, making the filtering system flexible rather than static.
3Measurement precision
If comprehensive transaction analysis is performed on all electronic services, then detection accuracy improves, but computational resource utilization increases significantly
Solution Approach 1:
The patent applies partial analysis by performing comprehensive machine learning model evaluation only on transactions that exceed initial risk thresholds or match specific high-risk patterns. For low-risk transactions, the system uses lightweight rule-based filtering or simplified scoring, performing exhaustive analysis only when necessary to maintain accuracy while reducing overall computational resource consumption.
Data Source
AI summary
Systems and methods for optimizing filters for processing electronic services between end users are disclosed. In an embodiment, a computer system accesses scores corresponding to historic user actions. A first cutoff value and second cutoff value are determined for branches of a tree. A precision score for branches of the tree is calculated based on a number of historic user actions having chargebacks in relation to a number of historic user actions captured by the cutoff values of the branch. The computer system identifies a branch having a greatest precision score. The computer system determines that a threshold, defined by the first cutoff value and the second cutoff value for the identified branch, when used in a processing rule for a user account, changes a performance metric by a threshold amount. The computer system generates a recommendation for the user account to adjust a filter for the processing rule to use the threshold.