基于蚁群系统的聚类算法研究

针对传统聚类算法在对复杂密集型数据集聚类时不能取得较好聚类结果的问题,利用进化聚类算法对复杂密集型数据集进行聚类,提出一种基于蚁群系统的聚类算法(clustering algorithm based on ant colony system,CAACS),利用蚂蚁在行进路径中释放信息素且追求浓信息素的原理来实现蚂蚁的随机搜索,并引入近邻函数值的概念来确定样本数据之间的相似性,通过蚂蚁在行走过程中不断建立样本数据之间的最相似连接来形成各个子连通图,各个子连通图中的样本数据构成一个类。实验采用随机产生的不规则数据集以及一系列合成的数据集将CAACS算法与DBSCAN算法(density-based spatial clustering of application with noise)及面向非规则非致密空间分布数据的蚁群聚类方法进行比较。实验结果表明CAACS算法对复杂密集型数据集能达到较好的聚类结果。

The research of a clustering algorithm based on the ant colony system

For solving the problem that traditional clustering algorithms can not get good results on clustering of complex data sets, a clustering algorithm based on the ant colony system is presented. The ant’s random search is realized based on the principle that the ant leaves pheromone on its path and pursuit concentrated pheromone.  The similarity among the data is determined based on the concept of neighboring function values.  Ants establish  connections between the most similar data in the process of walking,  draw the various sub-connected graph,  and the data in the same sub connected graph are given the same cluster number. Some experiments have been made to compare the results of the proposed CAACS algorithm with those  of the DBSCAN algorithm. The experiments are based on randomly generated irregular data sets and a series of synthetic data sets. The experimental results show that the CAACS algorithm can achieve good results on complex data sets.