基于极端梯度提升特征重要性的K-匿名特征选择  

K-Anonymous feature selection based on XGBoost feature importance

作　　者：王旭 万长胜[2] Wang Xu;Wan Changsheng(School of Cyber Science and Engineering,Southeast University,Jiangsu Nanjing 211189;Research Base of International Cyberspace Governance(Southeast University),Jiangsu Nanjing 211189)

机构地区：[1]东南大学网络空间安全学院,江苏南京211189 [2]网络空间国际治理研究基地(东南大学),江苏南京211189

出　　处：《网络空间安全》2020年第8期70-74,共5页Cyberspace Security

摘　　要：随着大数据时代的到来,数据在带来价值的同时,也损害了隐私。文章主要对关系型数据发布时的隐私保护模型进行了研究,针对K-匿名算法导致分类挖掘性能下降的问题,提出了基于极端梯度提升特征重要性的k-匿名特征选择算法XGB-KA,对满足k-匿名的特征子集进行了优选。通过与Greedy-Hamdist算法进行对比实验,验证了XGB-KA算法的有效性。With the advent of the era of big data,data brings value while compromising privacy.This paper mainly studies the privacy protection model when relational data is released.Aiming at the problem that the performance reduction of classification mining caused by the K-anonymity algorithm,a K-anonymity feature selection algorithm XGB-KA based on XGBoost feature importance is proposed to optimize the subset of features that satisfy k-anonymity.The effectiveness of XGB-KA algorithm is verified by a comparison with Greedy-Hamdist algorithm.

关 键 词：数据挖掘 隐私保护 K-匿名 特征选择 极端梯度提升 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]