近邻同步聚类模型与指数衰减加权同步聚类模型的比较与分析

The Comparison and Analysis Between Near NeighborSynchronization Clustering Model and Exponential DecayWeighted Synchronization Clustering Model

聚类是一种重要的数据分析与预处理技术。与传统的静态聚类分析方法相比,基于同步模型的聚类算法属于一种动态演化的聚类分析技术。先提出了应用到聚类中的两种指数衰减加权同步模型和一种δ近邻指数衰减加权同步模型。对前两种同步模型,提出了基于指数衰减加权同步模型的聚类算法;对后一种同步模型和已发表的扩展Kuramoto模型、Vicsek简化模型及Vicsek模型的一个线性版本,提出了基于近邻同步模型的聚类算法。然后比较分析了这些同步聚类模型的算法复杂度、性质及特点。在人工数据集和8个UCI数据集的仿真实验中,对这几种同步聚类模型在聚类精度、聚类速度等方面进行了适当的比较。最后对基于同步模型的聚类算法的发展进行了总结及展望。

Clustering is an important data analysis and preprocessing technology.Compared with the traditional static clustering analysis methods,the clustering algorithms based on synchronization models are a kind of dynamic evolutionary clustering analysis technique.Two exponential decay weighted synchronization models and aδnear neighbor exponential decay weighted synchronization model for clustering are proposed.For the first two synchronization models,a clustering algorithm based on exponential decay weighted synchronization model is proposed.For the latter synchronization model,the extended Kuramoto model,a Vicsek simplified model,and a linear version of the Vicsek model,a clustering algorithm based on near neighbor synchronization model is proposed.The algorithm complexity,properties,and characteristics of these synchronous clustering models are compared and analyzed.In the simulation experiments of some artificial data set and eight UCI data sets,these synchronization clustering models were compared in clustering accuracy and clustering speed.Finally,it summarizes the development of some synchronization clustering algorithms and presents the trends.