Synthetic Data Engine
Description
This AI based synthetic data generator provides synthetic data for a robust, large-scale dataset consistent with US Population Data Heuristics for testing and advanced analytics by dimensionalizing over 180 elements of socioeconomic data and over 300 dimensions, aging people, households and businesses over time, and tightly correlating the synthetic data to actual familial and socioeconomic attributes of US taxpayers.
Detailed example
AI is capable of outputting 3 tax years of IMF and BMF returns starting in Tax Year 2022. With each XML Schema Definition (XSD) that is released under the Modernized e-File (MeF), system implements each XSD version and provides full regression support for testing in successive years. Each XSD version has automated tests that run to ensure verifications can be run to reduce the likelihood of AI generated synthetic returns that might have data anomalies. Also, to help seed test systems with synthetic individuals and synthetic businesses, system can output the Data Master 1 (DM1) file simulating a feed from Social Security Administration (SSA) and Application for Employer Identification Number file (Form SS-4) containing transcribed form information so that the test system can be initialized with the appropriate backend reference data for validation of IMF and BMF entities, as well as the ability to validate incoming synthetic tax returns.
AI / analytics pattern
Classical/Predictive Machine Learning: Models trained on data to make predictions or classifications based on identified patterns or relationships.
Automation level / stage
c) Deployed – The use case is being actively authorized or utilized to support the functions or mission of an agency.
Expected benefit
Generate synthetic data for testing IRS tax processing systems and significantly reduce the risk of exposing taxpayer information while enabling more comprehensive testing of functional test cases. This system can now use to simulate fraud cases for testing negative and positive automated test cases.
Audit / financial statement impact
The output is not presumed to be high-impact and is not used as the principal basis for significant decisions/actions
Controls / human review
ATO: No; PIA: Not published
Data needed
Over 140 input feeds including US Census, US Bureau of Labor and many other sources, understand relations between variables and find the most relevant with regards to household income levels. divide the population into distinct groups defined by value ranges in sets of variables, generate new population for each group maintaining the frequency distribution for variables.