分布式异常检测中隐私保持问题研究

Privacy Preserving Outlier Detection

隐私保持是目前数据挖掘领域的一个重要方向,其目标是研究如何在不共享原始数据的条件下,获取准确的数据关系.本文采用现实的多方安全计算模式,结合数据干扰技术,提出了一种保持隐私的异常检测算法.该算法选择那些超出局部阈值距离的两点间距离及其序号进行通讯,为了保持原始数据的隐私,随机抽取一些正常范围内的两点间距离及其序号,在加入干扰后分散在异常信息中.理论分析表明该算法既提供了现实的数据隐私又保障了算法的性能.

Privacy preserving data mining has emerged to develop accurate models without sharing precise individual data records. Based on a practical secure multi-party computation model, an efficient algorithm for privacy preserving outlier detection with data perturbation techniques is proposed. The ID numbers of pairwise points whose distance exceeds the threshold is necessary to communicate among different sites. Besides, some pairwise points whose distance is within the threshold is chosen to hide private information. The algorithm maintains integrity and good privacy of the data sets of each party while keeping communication and computation cost low.