FuzzyRadar: visualization for understanding fuzzy clusters

Abstract

Fuzzy clustering assigns a membership degree (MD) on a datum to a cluster, which reflects real-world clustering scenarios but increases the complexity of understanding fuzzy clusters. Many studies have demonstrated that multidimensional visualization techniques are beneficial to fuzzy clusters analysis. However, empirically, no single existing visualization technique can support most analytical tasks featured by fuzzy clustering. This work proposes a new visualization called FuzzyRadar for understanding fuzzy clusters. Its basic idea is to combine the advantages of radial coordinate visualization (Radviz), which specializes in data-oriented analytical tasks, and parallel coordinate plot (PCP), which performs well in cluster-oriented analytical tasks. First, we adopt a compact and compounded layout to integrate Radviz and PCP into one visualization view. Then, we introduce a strip-edge-bundling method to reduce the visual cluster caused by PCP polylines and a histogram embedding method to facilitate the recognition of MD distribution. We also provide a group of additional visual encodings and a set of lightweight interactions. Finally, we use a case study to demonstrate the usability of FuzzyRadar and conduct a controlled quantitative evaluation to compare the performance of FuzzyRadar, Radviz, PCP, and scatterplot matrix. Result shows that FuzzyRadar supports all the seven examined analytical tasks well and presents a significant capability improvement compared with Radviz and PCP.