蚁群聚类算法中邻域变化规律的研究

Neighborhood's Variation in Ant Colony Clustering Algorithm

为了提高蚁群聚类LF算法的聚类效果,在对基本LF算法改进的基础上,算法迭代过程中又进一步采用邻域线性增大和线性减小两种不同的方法,通过UCI数据集Iris和Wine数据的验证,使用FM作为聚类效果的评判标准,发现采用邻域线性递减的方法在两种数据集上运行的结果都优于邻域递增和邻域保持不变的情形。邻域递减策略使算法在运行初期能够对待聚类数据粗略的分类,随着邻域的减小,蚁群对数据分类的粒度逐渐细化,算法迭代结束,达到最佳的聚类结果。

In order to improve the clustering effect of the LF algorithm, two different ways of neighborhood increases linearly and neighborhood decreases linearly were used in iterative process of the improvement of basic LF algorithm. Result o{ neighborhood linear decre ment method are better than the neighborhood increment and the neighborhood remain unchanged. The result was verified by Iris and Wine data in UCI data sets and FM was a criterion for clustering effect. Because clustering datas were rough classification through decreasing strat egy in the early running of the algorithm, with decrease gradually of neighborhood, the granularity of data classification was refined. The best results of clustering was achieved at the end of iterative.