摘要
生长、分级的自组织映射(Growing Hierarchical Self-Organizing Map,GHSOM)网络是自组织映射(Self-Organizing Map,SOM)网络的一种变体,它不仅具备了SOM网络可解释性强的优点,同时采用多层分级的结构,不需要预先定义好网络的结构和尺寸,解决了SOM由于竞争层神经元过多造成的训练时间过长的问题,却忽略了对样本向量各个分量在模型中重要性的分析,因此将一种新的输入模式分量和映射单元权向量之间的灰关联度引入到网络权值的调整过程中,对GHSOM算法进行了改进。运用于对电信客户行为的分类,从中获取了预测欺诈客户的关键指标,大大降低了输入样本的维度。结果显示,采用改进后的GHSOM算法降维后,分类正确率仍然可以达到94.59%。
The network of Growing Hierarchical Self-Organizing Map(GHSOM) is a kind of variety to the Self-Organizing Map (SOM) ,it not only can be explained clearly like SOM,but also its architecture grows both in a hierarchical and in a horizontal way,and has no use for fixed architecture that has to be defined a-priori ,so the problem leading by too much units to deal with is solved,while which ignores the analysis to the importance of each branch in the sample vectors.Thus an improved algorithm is proposed by importing a new gray relation degree between input pattern and the weight vector of the nodes which is trained in each SOM of GHSOM.Making use of this improved algorithm to realize telecom customer's behavior classification,we get key indexes to predict trick users from classification result,can greatly reduce the dimensions of input space.The result demonstrates classification accuracy can still reach 94.59 percent after taking use of the improved algorithm to reduce the dimensions.
出处
《计算机工程与应用》
CSCD
北大核心
2007年第11期193-196,共4页
Computer Engineering and Applications
基金
国家自然科学基金(the National Natural Science Foundation of China under Grant No.60375002)。