Efficient privacy-preserving classification construction model with differential privacy technology

作     者：Lin Zhang Yan Liu Ruchuan Wang Xiong Fu Qiaomin Lin 

作者机构：College of ComputerNanjing University of Posts and TelecommunicationsNanjing 210003China Jiangsu High Technology Research Key Laboratory for Wireless Sensor NetworksNanjing 210003China 

出 版 物：《Journal of Systems Engineering and Electronics》 (系统工程与电子技术（英文版）)

年 卷 期：2017年第28卷第1期

页      面：170-178页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 0839[工学-网络空间安全] 08[工学] 081201[工学-计算机系统结构] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：supported by the National Natural Science Foundation of China(61373017 61402241 61572260 61572261 61472192) the Scientific&Technological Support Project of Jiangsu Province(BE2014718 BE2015702) 

主　　题：Budget control Data mining Decision trees Trees (mathematics) 

摘      要：To address the problem of privacy disclosure during data mining, a new privacy-preserving decision tree classification construction model based on a differential privacy-protection mechanism is presented. An efficient classifier that uses feedback to add two types of noise via Laplace and exponential mechanisms to perturb the calculation results are introduced to the construction algorithm that provides a secure data access interface for users. Different split solutions for attributes of continuous and discrete values are provided and used to optimize the search scheme to reduce the error rate of the classifier. By choosing an available quality function with lower sensitivity for making decisions and improving the privacy budget allocation methods, the algorithm effectively resists malicious attacks that depend on the background knowledge. The potential problem of obtaining personal information by guessing unknown sensitive nodes of tree-type data is solved correspondingly. The better privacy preservation and accuracy of this new algorithm are shown by simulation experiments. © 1990-2011 Beijing Institute of Aerospace Information.