一种基于主成分分析的稀疏数据模式分类隐私保护算法  

A Pattern Classification Privacy Preserve Algorithm for Sparse Data Based on Primary Component Analysis

作　　者：原永滨[1,2] 杨静[1] 张健沛[1] 于旭[3] 

机构地区：[1]哈尔滨工程大学计算机科学与技术学院,哈尔滨150001 [2]福州大学电气工程与自动化学院,福州350108 [3]青岛科技大学信息科学与技术学院,青岛266001

出　　处：《科技导报》2014年第12期68-73,共6页Science & Technology Review

基　　金：国家自然科学基金项目(61370083;61073043;61073041);高等学校博士学科点专项科研基金(20112304110011;20122304110012)

摘　　要：模式分类过程涉及到对原始训练样本的学习,容易导致用户隐私的泄露。为了避免模式分类过程中的隐私泄露,同时又不影响模式分类算法的性能,提出一种基于主成分分析(PCA)的模式分类隐私保护算法。该算法利用PCA提取原始训练数据的主成分,并将原始训练样本集合转化为主成分的新样本集合,然后利用新样本集合进行分类学习。选用Adult数据集和KDD CUP 99数据集进行仿真实验,并采用正确率和召回率进行性能评价,结果表明,该隐私保护算法通过PCA提取原始数据特征属性的主成分,可避免原始属性的泄露,同时PCA在一定程度上可实现去噪,从而使分类器的分类性能优于原始数据集的分类性能。与已有算法比较,该隐私保护算法具有更好的模式分类精度和隐私保护性能。The pattern classification process involves the learning from the original training samples, which easily leads to privacy disclosure. In order to avoid the leaks of privacy in the pattern classification process and not to affect the performance of the algorithm, this paper proposes a pattern classification privacy preserve algorithm based on the primary component analysis （PCA）. This algorithm extracts the principal component of the original training data and converts the original training samples to new samples corresponding to the primary components. Then, a classification model is trained on the new samples. Experiments are carried out on the Adult data set and the KDD CUP 99 data set, and the precision and recall indexes are used to evaluate the proposed algorithm. It is shown that this algorithm can avoid the leakage of the original attributes through extracting the principal components of the feature attributes about the raw data. PCA can achieve de-noising to some extent, so that the classification performance on the classifier is better than that on the original data set. Therefore, compared with the existing algorithms, this algorithm has better pattern classification accuracy and privacy preserve performance.

关 键 词：主成分分析 模式分类 隐私保护算法 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]