Interpreting Black Box Models for Payment Authorization Decisions

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Black box machine learning models used for payment authorization decisions in electronic transactions lack transparency, making it difficult for users to understand which input data drive the authorization decisions, monitor the model, or remedy errors.

Innovation Solution

A computer-implemented method and system that interprets black box models by receiving an inquiry request for a payment transaction, querying a database to identify a subset of historical transactions with similar parameters but different authorization decisions, determining impact parameters by comparing transaction parameters, and generating an inquiry response based on these parameters.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If black box machine learning models are used for automated payment authorization decisions, then productivity and automation are improved, but transparency and interpretability deteriorate

Engineering Contradiction:

Improveautomated payment authorization decisionsVSAvoidtransparency of decision-making process

Core Design Contradiction:

ProductivityVSLoss of information

Solution Approach 1:

The patent introduces an intermediary system that sits between the black box model and the user. This intermediary queries the black box model with specific inputs, retrieves the decision, then反向 queries with modified inputs to determine which parameters drove the decision. This mediator translates the opaque black box output into interpretable information about parameter impacts without requiring changes to the original black box model.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system creates a copy or replica of the black box model's decision-making behavior by querying it with various inputs and recording outputs. This copying approach allows the system to reconstruct the decision logic indirectly through multiple queries, enabling interpretation without exposing the actual proprietary model internals. The copied behavior is then analyzed to determine parameter impacts.

Inventive Principle:

Principle #26Copying

2Device complexity

If black box models are used without interpretation, then device complexity is reduced, but the ability to monitor and remedy errors deteriorates

Engineering Contradiction:

Improvesystem structureVSAvoidmodel monitoring and error remediation

Core Design Contradiction:

Device complexityVSReliability

Solution Approach 1:

The patent implements a feedback mechanism where the system queries the black box model with original inputs, receives a decision, then queries again with modified inputs to observe how decisions change. This feedback loop provides information about which parameters most influence decisions, enabling monitoring of model behavior and identification of potential errors or biases without adding significant complexity to the system architecture.

Inventive Principle:

Principle #23Feedback

3Measurement precision

If all historical transactions are analyzed to interpret model decisions, then measurement precision is improved, but use of energy and computational resources worsen

Engineering Contradiction:

Improveaccuracy of impact parameter determinationVSAvoidcomputational resources for analysis

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent extracts only the necessary subset of information needed for interpretation. Instead of analyzing all historical transactions or all possible parameters, the system extracts specifically the parameter impacts by querying the model with targeted input variations. This extraction approach achieves sufficient measurement precision for understanding model decisions while minimizing computational resource consumption by focusing only on the most relevant information.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS20250139620A1System, Method, and Computer Program Product for Interpreting Black Box Models for Payment Authorization Decisions

Publication Date: 2025.05.01 VISA INTERNATIONAL SERVICE ASSOCIATION

AI summary

A computer-implemented method includes: receiving an inquiry request message identifying a first payment transaction having a first plurality of transaction parameters and a first authorization decision; querying a database including transaction data associated with a plurality of historical payment transactions to identify a subset of historical payment transactions, the transaction data including, for each of the plurality of historical payment transactions, a plurality of transaction parameters and an authorization decision, the subset of historical payment transactions including payment transactions having an authorization decision different from the first authorization decision and having a similarity score that satisfies a threshold; determining an impact parameter of the first plurality of transaction parameters by comparing the first plurality of transaction parameters with the plurality of transaction parameters associated with the plurality of historical payment transactions in the subset; and generating an inquiry response message based on the impact parameter.