Development of transformation based privacy preservation methods for data mining

Abstract

Data-mining is a task of discovering significant/salient patterns/rules/results from a set of large amount of data stored in databases, data warehouses or in other information repositories. Eventhough the focus on data-mining technology has been on the discovery of general patterns (not on any specific information regarding individuals) some data-mining applications may require access to individual s records having sensitive privacy data. Data containing structured information on individuals is referred to as micro-data. Abundance of recorded, personal information available in electronic form coupled with increasingly powerful data-mining tools, poses a threat to privacy and data security. The prime objective of this research is to find a solution to this problem. Eventhough, the identifying attributes are not published, some set of attributes in a released table (called quasi identifiers) may be linked with external data base leaking the sensitive data. To alleviate this problem, the so called K-anonymity and L-diversity principles and their improved versions have been used popularly in the earlier research works. But, all such methods suffer from proximity and divergence breach considerations. Also, the choice of generalization and diversity principles depends on the needs of underlying application, that is, application specific. Further, in such methods, if the level of privacy level is increased, information-loss also increases. And, these methods are successful only when the quasi-identifiers are explicitly identified with a surety of cent percent. Also, a set of organizations need to share their data so that datamining can be applied on the integrated data so as to get better rules/patterns/results. Normally, participating parties would like to accomplish mining tasks without disclosing sensitive information (in their databases) to other parties or to any third party. Various cryptographic techniques are used to preserve privacy of such collaborative data.