Information Technology Reference
In-Depth Information
Due to cost, the number of non-immersive 3D systems greatly outnumbers
immersive ones. However some immersive data mining projects are up and run-
ning. For example, a 3D virtual reality environment for data mining (3DVDM)
is discussed in Nagel et al. [12]. It uses 3D scatter plots with colored and shaped
icons to represent data points and summaries. Another virtual reality system for
data mining, TIDE [13], take a similar approach. Dive-On [14] displays take the
same form, but with the addition of OLAP style drill-down and roll-up in a im-
mersive environment. In [15] “Dynamic Visualization” is added to the 3DVDM
system, changing either the points displayed or their properties (color, size etc)
over time according to a fourth variable.
It seems clear that many research groups will be able to afford real 3D or large
screen/high-resolution displays in the near future. The question that interests
us most in this chapter is “what will be displayed on those screens?” Therefore,
the review below focuses on data visualization and visualization of data mining
discoveries.
4.2
Data Visualization
Approaches to Multivariate Data Visualization. The visualization of
multi-dimensional data is hard for two major reasons. First, the human visual
system is designed to perceive objects in three spatial dimensions (and one tem-
poral dimension). Secondly, most computer displays are normally only capable
of displaying information in two spatial dimensions (and one temporal dimen-
sion). So, to display objects with greater than 2 or 3 dimensions on a normal
screen, we need to do more than simply place an object at a particular posi-
tion in a 2-dimensional space. There are several approaches to the visualization
of multi-dimensional data that lacks natural spatial axes including: traditional
simple methods; object mapping; pixel based methods; dimension reduction; and
network visualization. These are summarized below with references to key or ex-
amplary work. A comprehensive overview of database visualization techniques
for exploratory analysis can be found in Daniel Keim and Mihael Ankerst's tu-
torial [22].
Simple Methods. Simple, traditional, non-computer based methods of graph-
ically presenting data include scatter plots, histograms and line graphs. The
computerization of these methods has had two major effects: firstly, these visu-
alizations are now quicker and easier to generate and secondly, they have been
extended into 3 dimensions. 3D effects may be obtained on paper by use of per-
spective, but a more effective method is to combine this with animated rotation.
Scatter plots represent the occurrence of data points with 2 numeric variables,
by plotting symbols on a 2D plane. A third (normally nominal) variable is often
used to determine which symbol is plotted. Computers have made it possible
to extend scatter plots to 3 dimensions. Rotating the plot, and using motion
parallax clues may obtain a 3D effect. In addition, more than 3 dimensions may
be displayed by collapsing several dimensions into one using techniques such as
projection pursuit or by using time as a fourth dimension.
Search WWH ::




Custom Search