Visual Analytics - Clustering (2)

In the second part of the clustering lecture, we discuss more clustering methods, in particular density-based methods, such as DB-SCAN. Hierarchical clustering methods create a whole hierarchy instead of a flat set of clusters. As major visualization of hierarchical clusters we describe dendograms. We also discuss how heterogeneous data that include numerical and categorical values can be clustered. Clustering, in general, is a non-supervised machine learning technique. It is however possible to incorporate expert knowledge in terms of constraints, e.g., certain elements should be in one cluster or must-not be in one cluster. We discuss some applications of such constraint-based clustering. We argue that there are several possible and valid variants of clustering data and therefore that multiple clusterings should be generated showing data from different perspectives. Finally, we briefly discuss the problem of clustering time-dependent data, where clusters may arise or disappear, they may merge or split – revealing changes over time. Chapters: 00:00 - Density-Based 08:01 - Hierarchical 27:41 - Clustering Categorical and Mixed Data 33:46 - Clustering with Constraints 39:47 - Combining Cluster Analysis and Regression 42:20 - Temporal Clustering 48:16 - Multiple Clustering 52:44 - Summary and References