I. LITERATURE STUDYPrivacy Preserving Data Mining (PPDM) was proposed simultaneously by D. Agrawal and CC Agrawal [1] and by Y. Lindell and B. Pinkas [5]. To solve this problem, researchers have since proposed various solutions that fall into two broad categories based on the level of privacy protection they offer. The first category of the Secure Multiparty Computation (SMC) approach provides the highest level of confidentiality; this allows mutually distrustful entities to mine their collective data without revealing anything except what can be inferred from an entity's own input and mining operation's sole output by Y. Lindell and B. Pinkas in [5], J. Vaidya and CWClifton. in [6]. In principle, any data mining algorithm can be implemented using O.Goldreich's generic SMC algorithms in [7]. However, these algorithms are extraordinarily expensive in practice and impractical for real-world use. To avoid the high computational cost, various solutions that are more efficient than generic SMC algorithms have been proposed for specific data mining tasks. Solutions for building decision trees on horizontally partitioned data were proposed by Y. Lindell and B. Pinkas in [5]. For vertically partitioned data, algorithms have been proposed to deal with association rule mining by J. Vaidya and CWClifton in [6], k-means clustering by J. Vaidya and C. Clifton in [8], and frequent model mining problems by AWC. Fu, RCW Wong and K. Wang in [9]. The work of B. Bhattacharjee, N. Abe, K. Goldman, B. Zadrozny, VR Chillakuru, M.del Carpio and C. Apte in [10] uses a secure coprocessor for collaborative data mining and analysis preserving confidentiality. The second category of partial information hiding approach concerns the middle of paper......that the coding system of WK Wong, DW Cheung, E. Hung, B. Kao and N. Mamoulis in [24] can be broken without using context-specific information. The success of the attacks in [25] mainly relies on the existence of unique, common and false elements, defined by WK Wong, DW Cheung, E. Hung, B. Kao and N. Mamoulis in [24]; our scheme does not create such elements, and the attacks of Y. Lindell and B. Pinkas in [5] are not applicable to our scheme. Tai et al. [9] assumes that the attacker knows the exact frequency of unique elements, just like we do..