基于网络社区模块结构的特征选择性能评价

Feature Selection Measurement Approach Based on Community Modularity Structure

利用网络社区模块结构作为特征选择的度量指标,给出了一种基于全局拓扑结构的特征选择性能评价方法。对一种基于免疫学原理的数据压缩和特征提取模型——人工免疫网络进行了验证,通过对数据特征提取前的抗原数据网络和特征提取后的记忆网络的网络社区模块结构的对比,达到对人工免疫网络(aiNET)的特征提取性能评价的目的。实验结果证实了人工免疫网络模型可以保持网络拓扑结构上的稳定性,验证了利用网络社区结构作为特征选择度量的合理性。

Taking community structure as measurement index for feature selection, a new feature selection measure approach based on modularity coefficient community structure is proposed. Artificial immune network is a type of competitive learning algorithm which is capable of extracting relevant features contained in dataset. It uses the ＂internal image＂ memory network to eliminate data redundancy and feature extraction. The new approach is used to analyze the community structure between the input pattern （antigen） and memory network. Experiment study shows the newly approach is proved reasonable and viable.