Attribute Selection in Correspondence Analysis of Incidence Matrices
本文研究关联矩阵的对应分析，提出一种方法选择对行差异贡献最大的列属性，并通过三个实例验证，适用于需要从高维频率数据中提取关键特征的经管研究。
SUMMARY An incidence matrix consists of frequencies for each of a number of attributes (in the columns) observed in each of a number of groups of observational units (in the rows). Three examples, from diverse applications, are provided. One way of analysing such matrices is by means of correspondence analysis, and this technique is illustrated on the three example data sets. In each case a geometrical configuration is derived in which the differences between the row profiles are best displayed. The close connection between the mathematics underlying both principal component analysis and correspondence analysis is then exploited, to obtain a method for selecting those columns of an incidence matrix which are primarily responsible for the highlighted row differences. This technique is applied to the three data sets. Some additional discussion concludes the paper.