Related Work

In recent years several research efforts have been directed at the display of large multivariate datasets.
 

Restructure the dataset to identify patterns: Wong et. al. [26]
Describes the construction of a multi-resolution display using wavelet approximations where the data size is reduced through repeated merging of neighboring points.

Advantage

• Identifies averages and details present in the dataset.

Disadvantage

• Requires the data to be ordered, making it useful only for datasets with a natural ordering, such as time-series data.

Let the characteristics of the dataset reveal itself: Wegman and Luo [23]
Suggests over-plotting translucent data points/lines so that sparse areas fade away while dense areas appear emphasized.

Disadvantage

• Relies on overlapping points/lines to identify clusters. Clusters without overlapping elements will not be visually emphasized.

Pixel-level Visualization Schemes: Keim et. al. [14]
Permits the display of a large number of records on a typical workstation screen.

Disadvantage

• The number of displayable records is dependent on the size of the display area. This limitation restricts the scalability of their method.
• Each pixel can only represent one variable making it difficult to convey the interactions among variables.

Visualizing hierarchical clusters using tree-map: Wills [25]
Extends upon the tree-map idea [20] by recursively subdividing the tree-map based on a dissimilarity measure. Their main purpose is to display the clustering end results, and in particular, the data partitions at a given dissimilarity value.

Disadvantage

• N-dimensional data characteristics of the cluster is not revealed.