一种基于自组织特征映射网络的聚类方法

针对传统聚类算法不能有效地处理大数据集和高维数据集的问题,提出了一种基于自组织特征映射网络的聚类方法。该方法能将任意维输入模式在输出层映射成一维或二维离散图形,并保持其拓扑结构不变,而且无需监督,能自动对输入模式进行聚类。给出了应用该方法的具体步骤和加速自组织过程的若干改进方法,通过仿真实验证明该算法的有效性。

Approach of clustering based on self-organizing feature map

The traditional clustering approaches cann't work effectively when the datasets are huge or high-dimensional. Due to this problem, a clustering approach based on self-organizing feature map is proposed. This approach can project the high-dimensional input space onto one-dimension or two-dimension discrete space. The learning process of clustering doesn't need supervision and it make input space cluster automatically. Furthermore, detailed algorithm steps and some advanced methods to accelerate the self-organizing process are presented. Finally, the experimental result demonstrates the effectiveness of this approach.