基于信息增益的模糊K-prototypes聚类算法

A fuzzy K-prototypes clustering algorithm based on information gain

K-prototypes聚类算法结合了K-means算法和K-modes算法,可用于分析混合属性的数据对象。传统的K-prototypes聚类算法在计算数据对象的相异度时,未考虑各个属性对于最终聚类结果的影响程度,而现实世界中,各属性的重要程度是不同的。使用了信息论中信息增益的计算方法,来获得各个属性的权值。在计算各属性的差异度时,乘以这些权值,从而可以获得更为准确的聚类结果。为了增加算法处理模糊问题的能力,本算法引用了模糊理论,从而使其具有较好的抗干扰能力和处理不确定性问题的能力。通过对四个UCI数据集的聚类分析实验,表明了本算法的有效性。

K-prototypes clustering algorithms combine K-means and K-modes to analyze mixed data objects. Classic K-prototypes clustering algorithms don＇t consider the effect degree of each attribute to the last clustering results when calculating the dissimilarity of data object. But in the real world,the im- portance of each attribute varies. In this paper we use information gain of the information theory to get the weight of each attribute. These weights are used to get a better clustering result when we calculate the dissimilarity. In order to improve the fuzzy ability, the proposed algorithm exploits the fuzzy theory to get a better capability for dealing with anti-noise and uncertain problems. Clustering experiments on four UCI data sets validate the effectiveness of our algorithm.