In this paper, we firstly state the secure three-vector product computation problem and develop protocols called tree-vector products to solve this problem. Our protocols are developed based on some basic protocols and their security properties is validated base on the composition theorem for the semi-honest model [9]. We then use the proposed protocols to address a focus problem of privacy preserving multivariate classification. | JOURNAL OF SCIENCE OF HNUE FIT. 2011 Vol. 56 pp. 72-80 PRIVACY PRESERVING MULTIVARIATE CLASSIFICATION BASED ON TREE-VECTOR PRODUCT PROTOCOL Luong The Dung VietNam Information Security Commission Tran Duc Su Academy of Cryptographic Technique E-mail thedungluong@ 1. Introduction Data mining has emerged as a significant technology for gaining knowledge from vast quantities of data 1 . Data mining technology allows us to analyze a personal data or a organizational data. However that creates threats to privacy this reason might cause an obstruction to data mining collaboration projects. For example two companies have a huge data set of records of their customers. These companies want to cooperatively conduct mining on their joint data set for their mutual benefit since this collaboration brings them results of mining which are more accurate than results of local data mining. However each company does not want to or does not have permission to disclose private information of other company s customers. The challenge then is whether the two companies above obtain results of mining while still preserving their data secrecy. Recently there has been growing focus on finding solutions to this problem 4 5 11 . In this paper we firstly state the secure three-vector product computation problem and develop protocols called tree-vector products to solve this problem. Our protocols are developed based on some basic protocols 3 4 7 8 and their security properties is validated base on the composition theorem for the semi-honest model 9 . We then use the proposed protocols to address a focus problem of privacy preserving multivariate classification. 2. Content . Background and problem statement . Problem statement Problem. Consider a data set D that has m observations n attributes X1 X2 . . . Xn and a class attribute Y . Where each Xi i 1 . . . n takes values as real numbers Y takes values as category data. The data set D is vertically 72 Privacy preserving .
