High Dimensional Data Visualization Analysis Based on Unsupervised Laplacian Score

Abstract

AbstractWith the rapid development of big data technology and information visualization technology, the concept of data visualization is constantly evolving and developing. As one of the classic high-dimensional data visualization methods, the parallel coordinate axis has excellent plane geometric characteristics. However, as the amount of data increases and the dimension of the data feature increases, the number of polylines on the finite plane of the parallel coordinate graph also increases. The crossing and occlusion of lines lead to serious visual redundancy and clutter. This project uses the feature distribution and feature axis arrangement on the parallel axis as the research entry point, and uses two unsupervised feature selection methods (Laplacian Score and SVD-Entropy) to re-arrange the features on the PCP axis to improve parallelism. Phenomena such as data disorder and clutter on the coordinate axis. Furthermore, we proposed a plane geometry optimization CLS algorithm by combining two unsupervised feature selection algorithms and the PCP axis radius coverage calculation method. The proposed algorithm conforms to people's perception characteristics of information and plane space representation, and can help people more quickly analyze and understand data.KeywordsHigh dimensional data visualizationParallel coordinate axisLaplacian scoreSVD-entropy