【作者】 覃华； 徐燕子； 张敏；

【Author】 QIN Hua,XU Yan-zi,ZHANG Min(College of Computer & Electronics Information, Guangxi University, Nanning, Guangxi, 530004, China)

【机构】 广西大学计算机与电子信息学院；

【摘要】 当数据样本特征复杂、类别多时,一般的蚁群聚类算法效果不好。分析其主要原因后,提出使用支持向量机的非线性映射函数把数据样本映射到核空间上,数据样本的特征在核空间中被重组和凸显,再在核空间上设计蚁群聚类算法,并用巢模板提高算法的稳定性和精确度。在UCI数据集上的实验结果显示:核空间上的巢模板蚁群聚类结果比较接近真实情况,能较好地处理特征复杂、类别多的数据集,其效果明显优于原空间上的聚类算法。更多还原

【Abstract】 If the data samples’ features are complex and have more categories, the ant clustering results are not satisfactory. After the analysis of the main reasons, an idea that maps the data samples to kernel space by SVM’s nonlinear mapping function is proposed. The data samples’ features are recombined and highlighted in kernel space. The ant clustering algorithm is designed in kernel space, then the nest template is been used to improve the algorithm’s stability and accuracy. Experimental results on UCI datasets show that the clustering results of nest template ant clustering algorithm in kernel space is closer to the truth. The algorithm can deal with datasets which are complex and have more categories and the result is better than the one in original space.更多还原