基于算法改进的ART2数据聚类方法研究

Research of data clustering based on improved algorithm of ART2

为提高聚类精度和产生更多类别相关信息,在分析了传统聚类方法对最初样本集过分依赖,不能动态适应样本空间变化,不能动态决定聚类数目等不足后,通过介绍其特性和优点提出应用自适应谐振神经网络(ART2)作为聚类算法;针对经典ART2模型的主观设置警戒参数、输出无组织等不足,提出基于改进算法的ART2模型用于聚类分析;通过自组织、迭代、加权等过程推导合理类别的聚类所需要的警戒参数,仿真实验证明了本算法的有效性.

In order to improve precision of cluster and produce more relevant information of classification, after analyzing the shortcomings of traditional clustering approaches, such as relying on initial sample sets excessively, inability of adapting sample space change and dynamically determining cluster＇s number, this paper suggests employing the adaptive resonance neutral network 2 （ART2） as a clustering algorithm in terms of the advantages and features of ART2. Then, on the basis of analyzing the shortcomings of classical ART2 model, such as subjectively setting vigilance parameter and unorganized output etc. , this paper presents an improved algorithm of ART2 model. The algorithm calculates the relevant vigilance parameter of clustering with reasonable cluster＇ s number through the progresses of self- organizing, iterating and weighting. And finally the effectiveness of the algorithm is demonstrated by simulation.