基于PMML的自组织神经网络元模型

THE PMML BASED META-MODEL OF SELF-ORGANIZING MAPS

Kohonen自组织特征映射网络SOM因其能够将高维数据映射为二维特征图而广泛应用于数据探索分析活动中。预测模型标记语言标准PMML是一个与平台及系统无关的数据挖掘模型表示语言,但其中并未包含SOM元模型的定义。通过对SOM模型的应用需求分析,提出了基于PMML的SOM元模型定义,可使模型生成与模型存储相分离,使用户在脱离模型生成系统的情况下进行模型的可视化及利用。

Self-Organizing Feature Maps （SOM）, as proposed by Kohonen, has been used as a tool for data exploratory analysis by mapping high-dimensional data into a two dimensional feature map. In the emerging standard PMML（ Predictive Model Markup Language） ,which is the platform and system independent representation of data mining models,there is no definition of the SOM meta-model. After the usage analysis of the SOM model ,we proposed an extension to the PMML model for SOM. The primary purpose of the PMML based meta-model of SOM is to separate model generation from model storage in order to enable users to visualize and utilize the SOM model independently of the system that generated the model.