A sanitization approach for big data with improved data utility

Abstract

The process of collaborative data mining may sometimes expose the sensitive patterns present inside the data which may be undesirable to the data owner. Sensitive Pattern Hiding (SPH) is a subfield of data mining that addresses this problem. However, most of the existing approaches used for hiding sensitive patterns cause high side-effect on non-sensitive patterns which in-turn reduces the utility of the sanitized dataset. Furthermore, most of them are sequential in nature and are not able to cope with massive amounts of data and often results in high execution time. To resolve these identified challenges of utility and non-feasibility, two parallelized approaches have been proposed named PGVIR and PHCR based on spark parallel computing framework which modifies the data such that no sensitive patterns can be extracted while maintaining the utility of the sanitized dataset. Experiments performed using benchmark dataset shows that PGVIR scales better and PHCR causes fewer side-effects to the data compared to the existing techniques.