Data Anonymization via Clustering and ILP Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data anonymization methods, particularly K-Anonymity approaches, face challenges in balancing data utility preservation with computational efficiency and information loss, especially in local recoding methods which incur high computational time and excessive generalization.
Innovation Solution
The proposed method employs an extended M-mode clustering technique and integer linear programming (ILP) to cluster records and generate patterns iteratively, using a weighted hierarchical distance metric and beta values to minimize information loss while achieving K-Anonymity, by considering attribute taxonomy trees for generalization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If local recoding is used to achieve K-Anonymity, then data utility is preserved, but computational time increases significantly
Solution Approach 1:
The patent divides the dataset into multiple clusters based on similarity metrics, and processes each cluster independently to generate anonymization patterns. This segmentation reduces the overall computational complexity while maintaining local recoding benefits within each cluster, thereby preserving data utility without incurring excessive computational time across the entire dataset.
Solution Approach 2:
The patent employs an iterative pattern generation process that dynamically adjusts generalization levels based on pattern support counts and information loss calculations. The algorithm adaptively refines patterns across multiple iterations, stopping when convergence criteria are met, thus optimizing the balance between data utility preservation and computational efficiency rather than applying static generalization rules.
2Loss of time
If global recoding is used to achieve K-Anonymity, then computational time is reduced, but data utility loss increases
Solution Approach 1:
The patent applies different generalization levels to different regions (clusters) of the data based on their specific characteristics and density. Instead of uniformly generalizing all records, the algorithm identifies local patterns within clusters and applies appropriate generalization only where necessary, thereby reducing overall data utility loss while maintaining computational efficiency through localized processing.
3Device complexity
If excessive generalization is applied to achieve K-Anonymity, then computational complexity is reduced, but information loss increases
Solution Approach 1:
The patent incorporates feedback mechanisms through pattern support counting and information loss calculation at each iteration. The algorithm monitors the quality of generated patterns and adjusts generalization levels accordingly, using the calculated information loss as feedback to determine whether further generalization is necessary. This feedback-driven approach prevents excessive generalization while managing computational complexity.
Solution Approach 2:
The patent dynamically changes generalization parameters (such as generalization level, pattern support thresholds) based on the characteristics of different data clusters and iteration progress. By adjusting these parameters adaptively rather than using fixed values, the algorithm reduces information loss while keeping computational complexity manageable through parameter optimization.
Data Source
AI summary
This disclosure relates generally to data anonymization using clustering techniques. In Typically, data anonymization using global recoding can overgeneralize the data. However, preservation of information while anonymization the data is of equal importance as obscuring the relevant information that can be used by the attackers. The disclosed method and system utilized attribute taxonomy tree for generalization to optimize the generalization of the records. The disclosed method uses clustering-based approach and after clustering, each cluster is solved independently using ILP model for K-Anonymization. The ILP model is solved by generalizing the value of the attributes. Sometimes, even after clustering the number of possible patterns is large, thus the disclosed method generates patterns on the fly during multiple iterations.