摘要： 针对聚类分析中隐私数据保护的问题，提出一种基于离散余弦变换矩阵的隐私数据保护方法(DCBT)，对随机选择的k个属性向量实施变换，直到所有属性都至少被变换一次且变换的次数达到初始设置值，选取隐私保护度最优的变换结果。实验结果表明，对于集中式数据，该方法能保持2点间距离不变，使数据较好地实现扭曲，保护隐私信息，对聚类结果基本没有影响。

关键词: 数据挖掘, 隐私保护, 聚类分析, 离散余弦变换

Abstract: Privacy preserving is an important direction for data mining research. This paper concentrates on the issue of protecting the underlying attribute values when sharing data for clustering and proposes a method called Discrete Cosine-Based Transformation(DCBT), random selects the k attributes and then distorts them with discrete cosine transformation. In the process of transformation, the goal is to find the proper chain of transform to satisfy the optimum privacy preserving requirement. For the centralized data, the experiments demonstrate that the method efficiently distorts attribute values, preserves privacy information and guarantees valid clustering results.

Key words: data mining, privacy preserving, clustering analysis, discrete cosine transform

中图分类号: 

• TP309