On utilizing dependence-based information to enhance micro-aggregation for secure statistical databases

摘要

We consider the micro-aggregation problem which involves partitioning a set of individual records in a micro-data file into a number of mutually exclusive and exhaustive groups. This problem, which seeks for the best partition of the micro-data file, is known to be NP-hard, and has been tackled using many heuristic solutions. In this paper, we would like to demonstrate that in the process of developing micro-aggregation techniques (MATs), it is expedient to incorporate information about the dependence between the random variables in the micro-data file. This can be achieved by pre-processing the micro-data before invoking any MAT, in order to extract the useful dependence information from the joint probability distribution of the variables in the micro-data file, and then accomplishing the micro-aggregation on the “maximally independent” variables—thus confirming the conjecture [A conjecture, which was recently proposed by Domingo-Ferrer et al. (IEEE Trans Knowl Data Eng 14(1):189–201, 2002), was that the phenomenon of micro-aggregation can be enhanced by incorporating dependence-based information between the random variables of the micro-data file by working with (i.e., selecting) the maximally independent variables. Domingo-Ferrer et al. have proposed to select one variable from among the set of highly correlated variables inferred via the correlation matrix of the micro-data file. In this paper, we demonstrate that this process can be automated, and that it is advantageous to select the “most independent variables” by using methods distinct from those involving the correlation matrix.] of Domingo-Ferrer et al. Our results, on real life and artificial data sets, show that including such information will enhance the process of determining how many variables are to be used, and which of them should be used in the micro-aggregation process.