Utility Independent Privacy Preserving Data Mining - Horizontally Partitioned Data

Abstract

Micro data is a valuable source of information for research. However, publishing data about individuals for research purposes, without revealing sensitive information, is an important problem. The main objective of privacy preserving data mining algorithms is to obtain accurate results/rules by analyzing the maximum possible amount of data without unintended information disclosure. Data sets for analysis may be in a centralized server or in a distributed environment. In a distributed environment, the data may be horizontally or vertically partitioned. We have developed a simple technique by which horizontally partitioned data can be used for any type of mining task without information loss. The partitioned sensitive data at 'm' different sites are transformed using a mapping table or graded grouping technique, depending on the data type. This transformed data set is given to a third party for analysis. This may not be a trusted party, but it is still allowed to perform mining operations on the data set and to release the results to all the 'm' parties. The results are interpreted among the 'm' parties involved in the data sharing. The experiments conducted on real data sets prove that our proposed simple transformation procedure preserves one hundred percent of the performance of any data mining algorithm as compared to the original data set while preserving privacy.