Review of Hierarchical Models for Data Clustering and Visualization

Real data often show some level of hierarchical structure and its complexity is likely to be underrepresented by a single low-dimensional visualization plot. Hierarchical models organize the data visualization at different levels, and their ultimate goal is displaying a representation of the entire data set at the top-level, perhaps revealing the presence of clusters, while allowing the lower levels of the hierarchy to display representations of the internal structure within each of the clusters found, providing the definition of lower level sets of subclusters which might not be apparent in the higher-level representation. Several unsupervised hierarchical models are reviewed, divided into two main categories: Heuristic Hierarchical Models, with a focus on Self-Organizing Maps, and Probabilistic Hierarchical Models, mainly based on Gaussian Mixture Models.