UNIT - V

DATA VISUALIZATION: Pixel-Oriented Visualization Techniques, Geometric Projection Visualization Tech-

niques, Icon-Based Visualization Techniques, Hierarchical Visualization Techniques, Visualizing Complex
Data and Relations.

Why is data visualization considered an art? Elaborate.

Data visualization helps gain insight into an information space by mapping data onto graphical primitives.
It provides a qualitative overview of large datasets, enabling easier interpretation and understanding. By
visualizing the data, one can search for patterns, trends, structures, irregularities, and relationships among
the data points. This process helps in identifying interesting regions and determining suitable parameters
for further quantitative analysis. Additionally, data visualization offers a visual proof of computer-generated
representations, making abstract data more tangible and understandable for decision-making. Data visu-
alization is an art because it transforms complex data into visually appealing and understandable stories,
combining technical skills with creative expression.
The following are the main categories of visualization techniques:

• Pixel-oriented visualization techniques

• Geometric projection visualization techniques
• Icon-based visualization techniques
• Hierarchical visualization techniques

In pixel-oriented techniques, data records can be ordered in a query-dependent way, such as sorting
them by their similarity to a point query in descending order. However, simply filling a window by laying
out the data records in a linear fashion may not work well for wide windows. The first pixel in a row is
far from the last pixel in the previous row, which creates a disjointed relationship in the layout. To solve
this problem, space-filling curves are used to fill the window efficiently. A space-filling curve, which covers
the entire n-dimensional unit hypercube, helps to address the layout issues by efficiently utilizing the space
in a 2D window. Furthermore, the circular segment technique offers a space-filling layout where dimension
windows are placed side by side in the shape of a circle, improving the comparison of dimensions. When
visualizing an m-dimensional data set, m windows are created on the screen, with each dimension’s values
mapped to pixels, and the colors of these pixels reflect the values of the corresponding dimensions. To save
space and show connections between multiple dimensions, space filling is often done in circular segments,
ensuring a compact and informative representation of the data.

Figure 1: Pixel-oriented visualization of four attributes by sorting all customers in income ascending order.

Geometric projection techniques are essential for visualizing high-dimensional data on a 2-D display,
addressing the challenge that pixel-oriented methods face in showing the distribution of data in multidimen-
sional space. A scatter plot is a basic geometric projection technique that displays 2-D data points using
Cartesian coordinates. Additional dimensions can be represented using colors or shapes, which helps in visu-
alizing relationships between dimensions. A 3-D scatter plot extends this idea by using three axes, and when

30
 Figure 2: Some frequently used 2-D space-filling curves.

Figure 3: The circle segment technique. (a) Representing a data record in circle segments. (b) Laying out
pixels in circle segments.

color is used, it can display up to 4-D data points. For data sets with more than four dimensions, scatter
plots become less effective. The scatter-plot matrix is an extension that visualizes pairwise combinations of
dimensions in a grid of 2-D scatter plots, making it useful for n-dimensional data sets. However, as dimen-
sionality increases, the scatter-plot matrix also becomes less effective. The parallel coordinates technique
is another popular method for visualizing high-dimensional data. It involves drawing n axes, one for each
dimension, and representing a data record as a polygonal line that intersects each axis at the correspond-
ing dimension value. While parallel coordinates are effective for higher-dimensional data, they can become
cluttered and hard to interpret when dealing with large datasets.
Icon-based visualization techniques, such as Chernoff faces and stick figures, use small icons to
represent multidimensional data. Chernoff faces, introduced by statistician Herman Chernoff in 1973, display
up to 18 variables as a cartoon human face. Different facial components, such as eyes, nose, mouth, and
ears, represent data values through their shape, size, placement, and orientation. For example, dimensions
can be mapped to features like eye size, mouth curvature, or eyebrow slant. Chernoff faces take advantage of
the human ability to recognize subtle differences in facial characteristics, making it easier to identify trends
and irregularities in the data. However, they have limitations, such as not displaying specific data values
and the varying importance of facial features depending on the mapping. Asymmetrical Chernoff faces, an
extension of the original technique, allow up to 36 dimensions by using both sides of the face.
Stick figure visualization, on the other hand, maps multidimensional data to five-piece stick figures. Two
dimensions are assigned to the x and y axes, while the remaining dimensions are represented by the angles
and lengths of the limbs. For example, census data could use stick figures to represent age and income on
the axes, with other dimensions like gender or education represented by the figure’s body. When data is
dense with respect to the axes, stick figures can create texture patterns that reveal data trends.
Other general techniques include shape coding, which uses different shapes to represent various types of
information, and color icons, which employ color to encode additional data. Tile bars, another technique,
use small icons to represent feature vectors and are commonly used in document retrieval systems to vi-
sually display information. These methods help in visualizing large datasets by condensing the data into

31
recognizable, meaningful icons, facilitating easier analysis and interpretation.

Figure 4: Chernoff faces: Each face represents an n-dimensional data point (n ≤ 18).

Hierarchical visualization techniquesare effective for visualizing high-dimensional data by parti-

tioning it into subspaces and presenting them in a hierarchical manner. One prominent example is the
“Worlds-within-Worlds” method, also known as n-Vision, which allows the visualization of a 6-dimensional
dataset by fixing certain dimensions and using 3D plots to represent subsets of the data. Users can interac-
tively adjust the location of the origin of the inner world and change the dimensions for both the inner and
outer worlds, enabling a dynamic exploration of the data. Another example of hierarchical visualization is
the tree-map, which displays data as a set of nested rectangles. Each rectangle represents a category, and
within each category, the data is further partitioned into smaller subcategories, making it useful for repre-
senting complex, hierarchical relationships, such as visualizing Google news stories categorized by topics.

How can data visualization help to show hierarchy? Explain with illustrations.

Data visualization can effectively represent hierarchical structures by organizing data in a way that
shows the relationships between different levels of the hierarchy. Several techniques can be used to illustrate
hierarchical data, making it easier to understand how items are nested or related. Some of the most common
methods include:
1. Tree Maps: A tree map visualizes hierarchical data as nested rectangles. Each rectangle represents
a node in the hierarchy, with its size reflecting the value or importance of that node. Sub-nodes or
categories are represented as smaller rectangles within the parent rectangle, providing a clear visual
representation of the relationships between different levels in the hierarchy.
Illustration: Imagine visualizing a company’s structure. The top-level rectangle could represent the
entire company, and smaller rectangles within it would represent departments. Each department can
be further divided into teams, and so on, with the size of the rectangles potentially showing the number
of employees or revenue for each unit.
2. Dendrograms: A dendrogram is a tree-like diagram that shows hierarchical relationships using
branches. Each branch splits into smaller branches, with the length or width of the branches often
representing the strength or significance of the relationships between nodes.
Illustration: In a biological classification system, a dendrogram can show how species are categorized
into kingdoms, phyla, and genera, with the branches illustrating the evolutionary distance between
different species.

32
 3. Sunburst Charts: A sunburst chart is a radial visualization where the hierarchy is shown in concentric
rings. The central circle represents the root node, and each subsequent ring represents a deeper level
of the hierarchy. This method is particularly useful for showing both the hierarchy and proportions
within each level.
Illustration: A sunburst chart could visualize a file system, with the root directory in the center and
folders or files as concentric rings around it. Each ring represents different levels of subdirectories or
files.
4. Indented Tree Lists: A simple and easy-to-read method, indented tree lists display hierarchical data
as a list with indentation indicating levels of hierarchy. Each level of the hierarchy is represented by a
list item indented further to the right, showing how elements are nested.
Illustration: A company’s organizational chart might be represented as an indented list, where the
CEO is at the top, and each successive level (e.g., department heads, managers) is indented beneath
their parent nodes.

By using these visualization techniques, hierarchical relationships can be effectively communicated, help-
ing users understand the structure of complex data sets and how different elements are connected.

Explain the challenges in visualizing complex data and relationships, and suggest suitable
mechanisms to address them.

Visualization techniques, initially used primarily for numeric data, have evolved to accommodate non-
numeric data, such as text and social networks. One popular method for visualizing non-numeric data is
the tag cloud, which displays user-generated tags with size or color indicating their importance. In a tag
cloud for a single item, the size of a tag reflects how frequently it is applied by different users, while for
multiple items, the size represents the tag’s popularity across all items. Additionally, visualizing complex
relationships among data entries poses its own challenges. For example, disease influence graphs represent
correlations between diseases, where nodes correspond to diseases, and their size indicates prevalence. Edges
between nodes show strong correlations, with the edge width representing the strength of these connections.
Beyond text data, methods like visualizing social networks also help in understanding complex relationships
within large datasets.
Evaluate tree maps as a hierarchical visualization technique.
Tree maps are an effective hierarchical visualization technique that displays data using nested rectangles,
with each rectangle representing a node in the hierarchy. The size of the rectangles often reflects the value
or importance of each node, and the color can be used to convey additional information such as categories
or performance metrics. One of the main advantages of tree maps is their compact representation, which
allows large datasets to be displayed efficiently in a limited space. They also provide a clear visualization
of hierarchical structures, with parent nodes represented by larger rectangles and child nodes nested within
them, making it easy to understand the relationships between different levels. Additionally, tree maps allow
for easy comparison of values, as the size of the rectangles helps users identify large or small categories and
trends. However, tree maps can be less effective for complex hierarchies, where deeply nested relationships
may make the visualization harder to interpret. They may also oversimplify the data, omitting nuanced
relationships, and the compact nature of the rectangles can make labeling difficult. Despite these challenges,
tree maps are particularly useful for visualizing organizational structures, financial data, or website con-
tent categorization, where hierarchical relationships are essential and size-based comparisons are important.
While they provide valuable insights for high-level overviews, tree maps are best used for datasets that are
not overly complex or deep in hierarchy.
Justify the use of geometric projection visualization techniques for multi-dimensional data.
Geometric projection visualization techniques are useful for understanding multi-dimensional data by
simplifying it into two or three dimensions, making it easier to analyze. They help identify patterns, trends,
or outliers and show relationships between different dimensions. Techniques like scatter plots and paral-
lel coordinates allow us to visualize multiple dimensions at once, with added features like color or size to

33
represent more data. These methods are interactive, allowing users to explore data by zooming, rotating,
or filtering. Overall, geometric projections make complex, high-dimensional data more accessible and help
users uncover valuable insights.

Explain the challenges in visualizing complex data and relationships, and suggest suitable
mechanisms to address them.

Visualizing complex data and relationships is challenging because of the high number of dimensions and
the intricate connections between data points. As the number of dimensions increases, it becomes harder to
display all the information clearly, leading to clutter and making it difficult to spot patterns. Additionally,
the relationships between variables may be non-linear, further complicating the visualization. To address
these challenges, techniques like dimensionality reduction (e.g., PCA, t-SNE) can simplify the data while
preserving its structure. Hierarchical methods, like tree maps or ”Worlds-within-Worlds,” break down data
into smaller, manageable sections. Interactive visualization allows users to explore data by zooming or
filtering, helping to reveal hidden patterns. Finally, visual encoding, such as using color, size, or shape, can
make the data more understandable by representing different dimensions in a clearer way. Together, these
methods make complex data easier to analyze and interpret.
Illustration of Chernoff Faces and Stick Figures in Data Visualization

1. Chernoff Faces Example

Imagine you are analyzing a dataset of 10 individuals with the following attributes: age, income, education
level, and number of years of work experience. Chernoff faces can be used to visualize the multidimensional
data by mapping each variable to a facial feature. For example:

• Eye size could represent the individual’s age.

• Mouth width could represent the income level.
• Nose length could represent the years of work experience.

• Eyebrow slant could represent the education level (higher degrees might have a more pronounced
eyebrow slant).
Each person in the dataset is represented by a face, and by comparing the faces, you can easily detect
patterns or outliers (e.g., people with very high income may have wider mouths). The faces allow for an
intuitive understanding of the relationships between the data dimensions without having to look at raw data.
2. Stick Figures Example
Let’s consider a dataset for employees in a company, where each employee has attributes such as age,
gender, income, and department. Stick figures can be used to map this data:
• Age and income can be mapped to the x and y axes, respectively.
• Gender might be represented by the figure’s limb length (e.g., male and female figures could have
different arm lengths).
• Department could be mapped to the angle of the figure’s limbs, indicating different departments for
different angles.
As data points are plotted on a stick figure grid, patterns in the data such as clustering of similar income
levels and ages can be easily observed through the positioning of the figures. For example, a group of
employees in the marketing department might be clustered together in the visualization, while employees
in the engineering department form another group. The stick figures help show the relationships between
multiple variables in a visually engaging way.
Reference: https://hanj.cs.illinois.edu/bk3/bk3s lidesindex.htm

34
stick figures.png

Figure 5: Census data represented using stick figures

35
Figure 6: Use of tree-maps to visualize Google news headline stories

36