Abstract: Based on the problem analysis on existing mixed data clustering algorithms, the graph-based relaxed clustering algorithm is used as the problem solving foundation. Introduced the gaussian kernel parameter self-adapting computing step, the graph-based relaxed clustering algorithm is improved on the foundation of the “Local Scale” idea, and the points distance calculating process is extended to the mixed data clustering. Moreover, depended on the previous improved steps and cluster ensemble technology, a new mixed data clustering algorithm is proposed. Lastly, the case experiments are completed, and the results prove that the new algorithm has high robustness and good cluster precision.

Key words: mixed attribute, relaxed clustering algorithm, self-adapting, cluster ensemble, robustness

摘要： 在分析现有混合属性数据聚类算法存在问题的基础上，选用基于图论的松弛聚类算法作为解决问题的“基石”；引入基于“Local Scale”思想的高斯核参数计算步骤，对基于图论的松弛聚类算法进行了自适应改进，并对其点对距离计算过程进行了面向混合属性的度量扩展。在上述两步改进的基础上，结合聚类集成技术，提出了一种新的混合属性数据聚类算法，并进行了实例验证，结果表明提出的算法具有较强的参数鲁棒性和较高的聚类精度。

关键词: 混合属性, 松弛聚类算法, 自适应, 聚类集成, 鲁棒性