Cluster aware Star Coordinates. (February 2018)

• Record Type: Journal Article
• Authors: Feng, Kang; Wang, Yunhai; Zhao, Ying; Fu, Chi-Wing; Cheng, Zhanglin; Chen, Baoquan
• Is Part Of: Journal of visual languages & computing. Volume 44(2018)
• Page Start: 28
• Page End: 38
• Publication Date: 2018-02
• Subjects: Dimensionality reduction -- Visual clustering -- Star coordinates -- High-dimensional data; Visual programming languages (Computer science) -- Periodicals; Visual programming (Computer science) -- Periodicals; Programming languages (Electronic computers) -- Semantics -- Periodicals; Langages de programmation visuelle -- Périodiques; Programmation visuelle -- Périodiques; Langages de programmation -- Sémantique -- Périodiques; Programming languages (Electronic computers) -- Semantics; Visual programming (Computer science); Visual programming languages (Computer science); Periodicals; Electronic journals; 005
• DOI: 10.1016/j.jvlc.2017.11.003
• Languages: English
• ISSNs: 1045-926X

Abstract

Abstract: Star coordinates is an important visualization tool for exploring high-dimensional data. By carefully manipulating the star-coordinate axes, users can obtain a good projection matrix to reveal the cluster structures in the high-dimensional data. However, finding a good projection matrix through axes manipulation is often a very tedious and trial-and-error process. This paper presents cluster aware star coordinates plot, which not only improves the efficiency of axes manipulation with higher cluster quality, but also enables users to learn the relations between cluster and data attributes. Based on the proposed approximated visual silhouette index, we introduce the silhouette index view, which interactively informs the user of the cluster quality of the projection. However, the user may still have no clue on how to manipulate the axes to improve the cluster quality. To resolve this issue, we propose a dimensionality reduction technique for visualization to progressively modify the projection matrix and improve the cluster results. Through this technique including a family of cluster-aware interactions, users can highlight important features of interest, such as points, clusters and dimensions, effectively investigate the change of cluster structures, and steer their relationship with the dimensions. In the end, we employ twelve high-dimensional data sets and demonstrate the effectiveness of our method through a series of experiments: comparison with state-of-the-art methods, interactive outlier detection, and exploration of cluster-dimension relationship.