Data Dimensionality Reduction Method Combining Intra-class and Inter-class Distance
Suzhi Zhang, Xiaoni Chen, Penghui Li, Qiang Cai · 2019
Traditional principal component analysis method has the problems of low computational efficiency and large memory consumption when faces high-dimension data. After the information entropy is introduced, although it reduces the memory and running time, but it cannot meet the requirements of the actual classification effect. In order to extract low-dimensional features with good discrimination ability, the idea of intra-class and inter-class distance is introduced, the principal component analysis data dimension reduction algorithm based on intra-class and inter-class distance is proposed. In this algorithm, firstly, the entropy of attribute information is calculated, and then the threshold value of information entropy is compared to it, which to realize the feature screening of data matrix. Then, the improved PCA algorithm based on the ideas of intra-class distance maximization and inter-class distance minimization is used to reduce the dimension of data. Finally, the data after dimension reduction is classified by KNN and SVM algorithms. Compared with the PCA, principal component analysis based on information entropy (E-PCA) and LDA algorithms, the experimental results show that the proposed algorithm can not only improve the result of dimension reduction, but also significantly improve the discrimination performance of low-dimensional data after dimension reduction.