Generative-Predictive Numeric Data Anonymization for Cloud Analytics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data anonymization techniques fail to effectively balance privacy preservation and data utility, particularly in cloud environments, as they either expose sensitive information or distort the underlying data structure and relationships, limiting their applicability and effectiveness in predictive analytics.
Innovation Solution
A framework that uses a generative model to preserve the underlying data structure and relationships by training on non-target columns and a predictive model to generate predicted target values, ensuring the anonymized data maintains the same structure and correlations as the original data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If data masking techniques are used to anonymize categorical data, then privacy protection is improved, but the technique is limited to only categorical types of data and cannot handle numeric data
Solution Approach 1:
The patent applies universality by developing a data anonymization system that can handle multiple data types (categorical, numeric, temporal) through a unified framework. The system uses different anonymization techniques for different data types while maintaining consistent privacy protection, thereby achieving multi-functionality and broad applicability across various data structures.
Solution Approach 2:
The patent employs parameter changes by dynamically selecting and adjusting anonymization parameters based on data type and characteristics. For numeric data, it uses techniques like random noise addition with controlled magnitude; for temporal data, it applies time-shifting transformations. This allows the system to adapt parameters to preserve privacy while maintaining data utility across different data types.
2Object-affected harmful factors
If data perturbation techniques are used to anonymize numeric data, then privacy protection is improved, but a small base leads to weak anonymization while a large base reduces the utility of the dataset
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting the magnitude and distribution of random noise added to numeric data. The system modifies perturbation parameters based on data characteristics, business requirements, and privacy needs, allowing fine-tuned control over the balance between anonymization strength and data utility without relying on fixed large bases.
Solution Approach 2:
The patent employs dynamics by making the anonymization process adaptive and flexible. The system dynamically adjusts perturbation parameters based on real-time requirements, allowing the same technique to produce different levels of anonymization and utility depending on the specific data and use case, rather than using a static large base approach.
3Object-affected harmful factors
If synthetic data generation techniques are used to create anonymized data, then privacy protection is improved, but the underlying data structure and key influencers of the original dataset are not preserved
Solution Approach 1:
The patent applies copying by creating synthetic data that replicates the statistical properties, distributions, and relationships of the original dataset. The system generates synthetic records that mirror the underlying data structure, correlations, and patterns while containing no identifiable information from the original data, thus preserving analytical value without compromising privacy.
Solution Approach 2:
The patent employs parameter changes by carefully controlling the generation parameters of synthetic data to match the statistical characteristics of original data. The system adjusts parameters such as distribution shapes, correlation structures, and data relationships to ensure the synthetic data maintains the same analytical properties as the original dataset while being completely anonymized.
4Object-affected harmful factors
If data shuffling techniques are used to anonymize data, then privacy protection is improved, but privacy information can be hacked from the shuffled data
Solution Approach 1:
The patent applies parameter changes by transforming data values through controlled modifications that prevent re-identification. Instead of simple shuffling, the system applies parameter transformations such as adding random noise, applying non-invertible functions, or transforming data through different distributions, making it impossible to reverse-engineer the original data even with knowledge of the anonymization process.
Data Source
AI summary
A system and method including receiving numeric data of a first dataset including a plurality of columns having numeric values with one of the plurality of columns specified as a target column; generating a trained generative model based on numeric values in non-target columns of the plurality of columns; generating a trained predictive model based on numeric values in non-target columns of the plurality of columns being input variables and the target column being a target variable; generating, by the trained generative model, a new set of numeric data for the non-target columns; generating predicted target values for the non-target columns by the trained predictive model using the new set of numeric data as an input to the predictive model; and generating anonymized numeric data for the first dataset by combining the new set of numeric data and the target column populated with the generated predicted target values.