Synthetic Data Testing for Sensitive Information Domains
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Financial and healthcare institutions face challenges in testing software updates and generating data sets that provide realistic testing results without exposing personally identifiable information (PII) or incurring excessive costs, as existing anonymization methods are imperfect and synthetic data sets may produce computationally erroneous outcomes.
Innovation Solution
A system and method that compares synthetic and active data sets using clustering algorithms to generate a testing data set that mimics the decisioning outcomes of the active data set, ensuring realistic testing without exposing PII, by selecting generated subjects and corresponding information that produce similar outcomes, thereby eliminating the need for actual PII in non-production environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If anonymization methods are applied to active data sets to create testing data sets, then PII protection is improved, but data realism and testing accuracy deteriorate
Solution Approach 1:
The patent creates synthetic copies of active data sets that replicate the statistical properties, relationships, and decisioning outcomes of real data without containing actual PII. These synthetic testing data sets are generated through algorithms that preserve the underlying data distributions and correlations, enabling accurate software testing while completely eliminating PII exposure risks.
2Object-affected harmful factors
If synthetic data sets are obtained from third parties for testing, then PII protection is improved, but data realism and decisioning outcome accuracy deteriorate
Solution Approach 1:
The patent performs preliminary analysis of the active data set to capture its statistical properties, relationships, and decisioning logic before generating synthetic data. By pre-processing the real data to extract these characteristics and then reproducing them in synthetic form, the system ensures that the generated testing data maintains high fidelity to the original data's decisioning outcomes while eliminating PII.
3Manufacturing precision
If active data sets are used directly for testing, then testing realism is improved, but PII exposure risk increases
Solution Approach 1:
The patent extracts only the essential statistical properties, relationships, and decisioning patterns from the active data set, separating these useful characteristics from the actual PII. By taking out just the necessary data attributes and their relationships while leaving behind the sensitive personal information, the system creates synthetic testing data that is realistic for testing purposes but completely free of PII exposure risks.
Data Source
AI summary
Computer-implemented methods, computer-readable media and apparatuses for testing executable code in sensitive-information domains are provided. A synthetic data set is received, where the synthetic data set includes generated subjects and corresponding information associated with the generated subjects. An active data set is received, where the active data set includes active subjects and corresponding information associated with the active subjects. The corresponding information associated with the active subjects includes sensitive information requiring privacy protection. The synthetic data set and the active data set are compared using an algorithm to identify the generated subjects that produce decisioning outcomes similar to decisioning outcomes of the active data set. A testing data set is generated by selecting the generated subjects identified using the algorithm and corresponding information associated with the identified generated subjects. The testing data set is output to a code-testing environment. Computer-executable code in the code-testing environment is tested using the testing data set.