基于模块性指标的特征提取方法

In order to solve the existing problem of feature extraction, a simple applied method is introduced. In this paper, modularity criterion in network analysis is adopted to evaluate the sensitivity of feature variable to classification, and the feature attribute which has less contribution is eliminated. A hierarchical clustering algorithm based on the modularity criterion optimization is used to assemble the data with the same attribute, the most representative one that is the center of feature cluster is obtained and accordingly the data redundancy is settled. The applications to benchmark data show that the new method can not only reduce data dimension and the size of training set significantly, but also reduce the calculation cost while increasing accuracy of classification.