Government Engineering College Sheohar

Chhatauna, Bisunpur, Block-Piprahi, Sheohar, Bihar Pin Code-843327

Department of Computer Science and Engineering (Data Science)

Assignment Submission Semester: 06

Sub: Data Visualization (Course

Code:153602)

Submitted by

Name: Vikash Kumar

Roll No: S-22530
Registration No: 22153145006

Submitted to: Dr. Shahadat Hussain

 Contents

1. Visualize Trends 02

2. Geospatial data and uncertainty 05

3. Visualization of Networks and Trees 06

4. Visualizing Multidimensional data 09

5. Data Reduction – Reduce Items and Attributes 13

6. The principles of proportional link 15

7. Handling overlapping points 17

8. Balance the data and context 19

9. Understanding the most commonly used Image file formats 21

10. Choosing the right visualization software 23

1|Page
 1. Visualize Trends

"Visualizing trends" generally refers to data visualization, the process of transforming quantitative
information into graphical representations to make patterns and correlations easier to identify.
The most common and effective way to visualize trends over time is by using line charts. Other useful
charts include bar charts, scatter plots, and area charts, each suited to different types of data and insights.

 Common Types of Visualizations for Trends:

Different chart types serve specific purposes when visualizing data trends.
 Line Charts: The go-to option for showing how values change over a continuous period (time-
series data). They help you spot increases, decreases, or stability at a glance.

 Bar Charts: Used to compare values across

different categories. They can show changes in
rank or magnitude over time when used in series
or in specific formats like waterfall charts.

2|Page
  Scatter Plots: Ideal for exploring relationships and correlations between two different variables.
Each data point is represented by a dot, helping to uncover hidden patterns or outliers.

 Area Charts/Streamgraphs: Similar to line charts, these fill the area beneath the line to
emphasize volume and show how different categories contribute to a whole over time.

3|Page
  Dashboards: These combine multiple visualizations into a single, interactive interface, offering
real-time insights for comprehensive analysis of various trends simultaneously.

 Tools and Resources:

Several tools can help you visualize trends:
 Analytics Tools: Modern business intelligence tools (like Tableau or Qlik) provide user-friendly
interfaces and extensive libraries of chart types to simplify the process of visualizing data.

 Coding Libraries: For data science professionals, programming libraries such as Python's
Matplotlib or R's ggplot2 offer powerful, customizable visualization capabilities.

 Public Data Platforms: You can explore existing data visualizations or use the Google
Trends platform to track what is currently spiking in searches around the world.

4|Page
 2. Geospatial data and uncertainty

Uncertainty in geospatial data is the inherent degree of doubt or error associated with the location and
attributes of geographic phenomena. This uncertainty is unavoidable and impacts the reliability of spatial
analysis and decision-making in fields like urban planning, environmental management, and public
health.
 Sources and Types of Uncertainty
Uncertainty can arise at every stage of the data lifecycle, from collection to analysis. Key sources
include:
 Measurement Errors: Inaccuracies during data collection (e.g., GPS positional errors, sensor
limitations, or human mistakes during manual entry).
 Natural Variability: The real world is complex and heterogeneous (e.g., soil properties or land
cover can change gradually), making crisp boundaries and classifications difficult to define
perfectly.
 Data Processing: Errors introduced when data is manipulated, transformed, or interpolated to fill
gaps (e.g., converting vector data to raster, or resampling to a different scale).
 Incompleteness/Age of Data: Missing information, an incomplete geographic area, or outdated
data introduce uncertainty because the map no longer reflects the current ground truth.
 Vagueness and Ambiguity: Issues arising from the subjective interpretation of definitions or
classification systems (e.g., "urban area" might be defined differently by various observers).

 Management and Communication

Managing uncertainty is crucial for ensuring that geospatial analysis is fit for its intended purpose.
Strategies include:
 Quantification: Using statistical methods like probability theory, error propagation models, and
Monte Carlo simulations to quantify the potential range of error.
 Metadata Documentation: Providing comprehensive documentation (metadata) about the data's
origin, quality, and processing steps helps users understand its limitations.
 Visualization: Using cartographic techniques to visually represent uncertainty to users. This can
involve color coding, transparency, error bars, or fuzzy boundaries.
 Sensitivity Analysis: Assessing how changes in input data or assumptions influence the final
results, which helps identify critical sources of uncertainty.
 Uncertainty Absorption: Accepting that some level of uncertainty is unavoidable or too costly to
reduce further, and making informed decisions while acknowledging the associated risk.

5|Page
 3. Visualization of Networks and Trees

Network and tree visualizations use nodes (entities or data points) and edges (connections or
relationships) to represent complex relational data. These techniques are powerful for uncovering
structures like social networks, organizational charts, biological systems, and file directories.

 Visualization of Networks
Networks (or graphs) represent interconnected entities where relationships can be complex, undirected, or
weighted.

Common Techniques:

 Node-Link Diagrams: The most common method, using points (nodes) and lines (edges). Various
layout algorithms are used to position nodes effectively:

o Force-Directed Layouts: Simulate physical forces, pushing connected nodes together and
unrelated nodes apart to reveal clusters and central nodes.

o Circular Layouts: Arrange nodes in a circle, useful for highlighting connections around the
perimeter.

o Geospatial Layouts: Place nodes on a map based on their geographic location to visualize spatial
relationships.

 Matrix Views (Adjacency Matrix): An

alternative to node-link diagrams, where a grid
represents connections. This is good for large,
dense networks to avoid edge clutter but makes
following paths more difficult.

6|Page
 Visualization of Trees

Trees are a specific type of network that depicts a hierarchical structure, starting from a single
"root" node that branches into child nodes without forming cycles or loops.

Common Techniques:

 Node-Link Diagrams (Hierarchical Layout): Organize nodes in layers (top-down or left-

right) based on their depth from the root, often seen in organizational charts or file system
views.

 Radial Trees: Position the root in the center, with child nodes radiating outwards in
concentric circles. This layout can use space efficiently for balanced trees.

 Treemaps: Use nested rectangles to represent the hierarchy, where the size and color of
each rectangle can represent quantitative attributes (e.g., file size in a directory). This is a
space-filling approach.

 Sunburst Diagrams: Similar to treemaps but use a radial, ring-based layout to show
hierarchical proportions.

7|Page
Key Considerations

Choosing the right visualization depends on the data's properties and the goal of the analysis:

 Clarity and Readability: Effective visualizations minimize edge crossings and node overlaps to
prevent clutter, especially with large datasets.

 Interactivity: Tools often incorporate features like zooming, filtering, or collapsible nodes to
manage complexity and allow for detailed exploration.

 Encoding Attributes: Visual attributes like color, size, and shape can be used to encode additional
data about the nodes or edges (e.g., importance, type, or strength of relationship).

8|Page
 4. Visualizing Multidimensional data

Visualizing multidimensional data involves techniques that represent datasets with more than two
variables (dimensions or attributes) to help identify patterns, correlations, and outliers that are not visible
in simple 2D charts. Since we are limited by the two dimensions of a screen, these methods cleverly use
color, size, shape, or space-filling techniques, or reduce the data's dimensionality.

Here are several common techniques for visualizing multidimensional data:

1. Parallel Coordinates Plots

Each dimension is represented by a vertical, parallel axis. A single data point is drawn as a polyline that
crosses each axis at a position proportional to its value for that dimension.

 Best for: Identifying patterns, correlations between adjacent variables, and spotting outliers in
datasets with a high number of dimensions.

 Challenge: Can become visually cluttered and difficult to read with a large number of data points.

9|Page
2. Scatter Plot Matrices (SPLOM)

This technique creates a grid of standard 2D scatter plots, displaying every possible pairwise
combination of variables in the dataset.

 Best for: Quickly exploring relationships and correlations between pairs of variables and
detecting anomalies.

 Challenge: The total number of plots grows rapidly with more dimensions (𝑛2 plots),
making it difficult to find patterns involving more than two variables simultaneously.

10 | P a g e
3. Using Visual Channels (Glyphs/Bubble Charts)
For datasets with up to 5 or 6 dimensions, additional variables can be mapped onto existing 2D or 3D
plots using different visual attributes:

 Color/Hue: Represents an additional continuous or categorical variable.

 Size: Used to encode a quantitative value, as seen in a bubble chart (which uses X-axis, Y-axis, and
size).

 Shape/Symbol: Distinguishes different categories of data points.

 Glyphs: Individual icons (like stars or Chernoff faces) are placed on a 2D plane, with each part of
the glyph representing a different dimension's value.
4. Dimensionality Reduction Techniques
For datasets with dozens or hundreds of dimensions, algorithms are used to project the data into a lower,
2D or 3D visual space while preserving as much variance or structure as possible.

 Principal Component Analysis (PCA): A linear method that identifies the directions (principal
components) in which the data varies the most.

 t-SNE and UMAP: Non-linear techniques well-suited for visualizing clusters and local structure in
high-dimensional data.

11 | P a g e
Tools for Multidimensional Visualization
Various tools are available for creating these visualizations, ranging from programming libraries to
business intelligence platforms:

 Libraries: Python's Matplotlib, Seaborn, and Plotly, or R's ggplot2 offer extensive customization.

 Software: Business intelligence tools like Tableau and Power BI provide interactive, user-friendly
interfaces.

12 | P a g e
 5. Data Reduction – Reduce Items and Attributes

Data reduction techniques minimize the volume of a dataset to increase storage efficiency, speed up
analysis, and improve model performance, all while striving to maintain the integrity of the original
information. This involves two primary strategies: reducing the number of data points (items/rows) and
reducing the number of variables (attributes/columns).

1. Reducing Items/Rows (Numerosity Reduction)

Numerosity reduction techniques focus on decreasing the number of records or observations in the dataset
while preserving general patterns and trends.

 Sampling: Instead of analyzing the entire large dataset, a smaller, representative subset (sample)
is selected. Common methods include simple random sampling, cluster sampling (sampling
groups of records), and stratified sampling (ensuring representation from different subgroups or
"strata").

 Clustering: Data points are partitioned into groups (clusters) based on similarity. The reduced
dataset can then be represented by the cluster summaries (e.g., the centroid or average point of
each cluster) rather than every individual data point.

 Data Cube Aggregation: Data is summarized by applying aggregation functions (like SUM,
COUNT, AVERAGE) across different dimensions in a data cube structure. For example,
quarterly sales data might be aggregated into annual sales totals, reducing the number of rows
from twelve per year to just one.

 Histograms/Binning: Continuous data is partitioned into disjoint intervals or "bins" (e.g., age
groups like 20-30, 30-40, etc.). The raw values are replaced by the bin label or a representative
value (mean/median), reducing the number of distinct values.

2. Reducing Attributes/Columns (Dimensionality Reduction)

Dimensionality reduction techniques aim to reduce the number of features or variables by eliminating
redundant or irrelevant attributes or by creating new, smaller sets of variables that capture the essential
information.

 Feature/Attribute Subset Selection: This method identifies and removes irrelevant or weakly
relevant attributes from the original set. Methods include:

13 | P a g e
 o Stepwise Selection: Iteratively adding (forward selection) or removing (backward
elimination) attributes based on a statistical measure of significance.

o Decision Tree Induction: Attributes that are not used to build the decision tree model
are assumed irrelevant and discarded.

 Feature Extraction/Projection: This involves transforming the data from the high-dimensional
space into a new space with fewer dimensions.

o Principal Component Analysis (PCA): A widely used technique that creates a smaller
set of uncorrelated principal components that capture most of the variance in the original
data. The weaker components (with low variance) are dropped.

o Wavelet Transforms: This mathematical technique (often used in image/signal

processing) transforms the data vector into wavelet coefficients; a compressed
approximation is retained by keeping only the strongest coefficients.

Trade-offs
Data reduction offers significant benefits in efficiency and storage optimization, but it's important to be
aware of the inherent trade-offs, primarily the potential for information loss. The goal is to achieve a
reduced representation that still produces the same or nearly the same analytical results as the original
large dataset.

14 | P a g e
 6. The principles of proportional link

The core principle is the Principle of Proportional Ink, which states that the amount of ink used to
represent a numerical value in a graphic should be directly proportional to that value itself. This ensures
clarity and prevents misleading interpretations of the data.

Key Principles of Proportional Link

This principle, inspired by Edward Tufte's work on graphical integrity, is a foundational rule in effective
data visualization.

 Direct Proportionality: The visual extent of a graphical element must accurately reflect the
magnitude of the data it represents. If one value is twice as large as another, its corresponding
visual element (e.g., a bar, a circle's area, or a pie slice) should be twice as large.

 Avoid Truncated Axes: For visualizations that use shaded areas or bar lengths to encode values
(like bar charts), the axis must start at zero. Starting an axis at an arbitrary non-zero value can
visually exaggerate differences, violating the principle by making the length of the bar represent
the difference from the baseline, not the total value.

 Consistency in Mapping: When using size as a visual channel, ensure the correct property is
mapped proportionally. For instance, in a bubble chart, the area of the bubble should be
proportional to the value, not the radius or diameter, as human perception is better at judging
distance than area.

 Context for Line Graphs: The principle of proportional ink primarily applies to shaded areas or
lengths. For line graphs, which primarily use position to indicate quantities, the axis does not
necessarily need to start at zero. The axis should be scaled to best highlight trends in the data.
However, if the area below the line is filled with ink, the axis must start at zero to maintain
proportionality.

 Minimize Non-Data Ink: Every visual element (ink) in a chart should contribute meaningfully to
understanding the data. Avoid unnecessary visual clutter, gratuitous 3D effects, or decorative
elements that obscure the meaning or distort proportions.

Common Violations

15 | P a g e
Violations of the proportional ink principle can be found in various contexts, particularly in the popular
press:

 3D Charts: Often distort perspective, making it difficult to accurately compare the lengths or
areas of different data points.

 Inconsistent Scaling: Using different scales or units within the same graphic without clear
indication can mislead the viewer.

 Misleading Symbols: Using symbols where size represents a value but scaling the radius instead
of the area leads to misinterpretation.

16 | P a g e
 7. Handling overlapping points
Overlapping points (also known as overplotting or occlusion) occur when multiple data points are
plotted at the same or very similar locations on a graph, obscuring the true data density and hidden
patterns. This is a common issue in scatter plots, especially with large datasets or data with low precision
(rounded values).

Here are effective techniques for handling overlapping points:

Adjusting Point Appearance

 Partial Transparency (Alpha Blending): Making the individual points semi-transparent is one
of the most effective methods. Areas with a higher concentration of points will appear darker or
more saturated, visually indicating data density while allowing individual points to be seen.

 Reduce Point Size: Simply making the markers smaller can reduce the amount of overlap,
allowing more individual points to be visible, although there is a limit to how small they can be.

 Change Point Shape/Color: When visualizing different data series, using different shapes or a
combination of shapes and colors can help distinguish overlapping points. This works best for a
small number of distinct categories.

Modifying Point Positions

 Jittering: For data points with identical or very close values (e.g., rounded data), a small amount
of random noise or displacement can be added to their positions in the x or y direction. This
spreads the points out so their density becomes apparent, making it clear that multiple
observations exist at that location.

 Stacking (Gather Plots): A more organized approach to jittering involves stacking overlapping
points in an orderly way (e.g., in a small grid or along a specific axis) rather than randomly
displacing them. This provides the benefit of no overlap while maintaining a clear count of
points.

17 | P a g e
Changing the Visualization Type

When the number of data points is very large and the above methods are insufficient, alternative chart
types that aggregate data are more suitable:

 2D Histograms (Hexbin Plots): The plot area is divided into a grid (usually hexagonal or
rectangular), and the color or shading of each bin represents the count or density of data points
falling within that area.

 Contour Plots (Density Plots): This method estimates the data point density across the plot area
and uses contour lines or shaded regions to delineate areas of similar density. This is good for
showing overall distributions and trends.

 Sampling: If the dataset is massive, a representative random subset can be used for the
visualization. While this is a simple approach, it may leave out important outliers or interesting
smaller patterns.

 Small Multiples: The data can be broken down into separate plots (small multiples) for different
subgroups or variables, allowing for easier comparisons without the clutter of a single, dense
visualization.

Ultimately, the best approach depends on the data size, the level of overlap, and the message you want to
convey.

18 | P a g e
 8. Balance the data and context

Balancing data and context involves ensuring that a visualization provides enough information for
accurate interpretation and decision-making without overwhelming the audience with irrelevant details.
The goal is to maximize the data-ink ratio within reason, meaning most ink should be dedicated to data
representation, but surrounding elements are vital for clarity and structure.

The Importance of Context

Raw data alone is often meaningless or can be misinterpreted. Context provides the "why" and the "so
what," transforming numbers into actionable insights. Non-data elements that provide context include:

 Titles and Labels: Clear, concise titles and labeled axes tell the viewer exactly what they are
looking at and the units of measurement.

 Annotations: Text callouts can highlight specific data points, explain anomalies, or point out key
insights that support a narrative.

 Reference Lines/Benchmarks: Including historical averages, targets, or comparison lines (e.g.,

an x=y line for paired data) helps the audience assess performance or relationships instantly.

 Metadata and Sources: Documentation about where the data came from, when it was collected,
and any limitations builds trust and allows for verification.

Techniques for Balance

Finding the right balance requires thoughtful design choices that guide the audience's attention to the data
while providing necessary scaffolding.

 Minimize Non-Data Ink (The "Within Reason" Rule): Use light, neutral colors (like light
gray) for grid lines, axis lines, and backgrounds so they recede visually, preventing them from
competing with the actual data points.

 Progressive Disclosure: Start with a high-level overview or summary. Provide the option for
users to "drill down" into more granular details or access supplementary information via tooltips,
filters, or separate detailed reports if needed. This avoids overwhelming the initial audience.

19 | P a g e
  Strategic Use of Color and Size: Use strong or bold colors and size variations to draw attention
to the most important data points or key insights, while using more muted tones for contextual
elements or less critical data.

 Audience-Centric Design: Tailor the level of detail to the audience's expertise and needs. C-
suite executives might need a high-level, narrative-driven visual, while data analysts might prefer
more detailed, interactive dashboards for exploration.

 Small Multiples: When comparing data across many categories, use a series of small, identical
charts (small multiples). The consistent axes and layout serve as inherent context, making the
data differences easy to compare without needing extensive annotations on each panel.

20 | P a g e
 9. Understanding the most commonly used Image file formats

Understanding common image file formats helps in selecting the right one based on needs like quality,
file size, transparency, and intended use (web or print). The most common formats are categorized as
either raster (pixel-based) or vector (mathematical path-based) graphics.

Common Raster Formats (Pixel-Based)

Raster images are composed of a grid of pixels and are best for complex, detailed images like
photographs. They lose quality and become pixelated if scaled up significantly.

Format Best Use Cases Compression Type Transparency Support

JPEG/ Photographs, web images, email Lossy (discards some data to No

JPG sharing reduce file size)

PNG Logos, illustrations, screenshots, Lossless (retains all data) Yes (supports variable
graphics with text transparency)

GIF Simple animations, web graphics with Lossless (but limited color Yes (simple
limited colors (max 256) palette) transparency)

TIFF Professional printing, high-resolution Lossless (optional) Yes

scans, archiving artwork

WebP Modern web images (aims to replace Lossy and Lossless Yes
JPEG, PNG, GIF)

Common Vector Formats (Scalable)

Vector images use mathematical equations to define lines and shapes, so they can be scaled
infinitely without losing quality or becoming pixelated.

 SVG (Scalable Vector Graphics): The standard for web-based vector graphics, ideal for
logos, icons, diagrams, and illustrations. It is an open standard and can be manipulated
with code.

21 | P a g e
  EPS (Encapsulated PostScript): A format often used in professional printing for vector
graphics and logos that need high precision.

 AI (Adobe Illustrator Artwork): Adobe's proprietary format for complex vector

illustrations, widely used by graphic designers.

When to Use Which Format

 For the Web: Use JPEG for photographs to balance quality and small file size.
Use PNG for logos or graphics requiring transparency or sharp lines. Use GIF for
simple, short animations. Consider newer formats like WebP for better compression and
features across the board.

 For Printing: Use TIFF or PDF for high-quality, professional prints where detail is
paramount. Ensure images are set to a high resolution (e.g., 300 DPI) and use the CMYK
color profile when possible.

 For Editing: Use a lossless format like PNG or the software's native format
(like PSD for Photoshop) to avoid quality degradation with repeated saving and editing.

22 | P a g e
 10. Choosing the right visualization software

Choosing the right visualization software depends heavily on your skill level, specific use case, data
complexity, budget, and integration needs. The available options range from simple spreadsheet programs
to complex business intelligence (BI) platforms and coding libraries.

Here is a guide to selecting the best software category for your needs:

1. Business Intelligence (BI) Platforms

BI tools are designed for interactive dashboards, data exploration, and sharing insights across an
organization. They typically require minimal coding and offer rich visual interfaces.

 Best for: Business dashboards, real-time data monitoring, data exploration, enterprise-wide
reporting.

 Top Choices:

o Tableau: Highly interactive, excellent visual aesthetics, and very flexible for exploring
complex datasets. Offers a free Public version for learning or public data.

o Power BI: Strong integration with Microsoft ecosystem (Excel, Azure), often more
budget-friendly, and very capable for standard business reporting.

o Qlik Sense: Known for its associative data engine, which allows users to explore data
dynamically and find hidden insights quickly.

2. Spreadsheet Software

These are the most accessible and widely used tools for basic data analysis and visualization.

 Best for: Quick analysis, small datasets, simple charts (bars, lines, pies), everyday data
management.

 Top Choices:

o Microsoft Excel: Ubiquitous and capable of creating standard charts, pivot tables,
and basic dashboards.

23 | P a g e
 o Google Sheets: Excellent for collaborative, cloud-based data handling and simple
visualizations.

3. Programming Libraries

For data scientists and developers who need high levels of customization, automation, and
statistical rigor, programming libraries offer maximum control.

 Best for: Statistical analysis, predictive modeling visualizations, highly customized

interactive web visualizations, automation.

 Top Choices:

o Python (Matplotlib, Seaborn, Plotly, Altair): A powerful ecosystem covering

static plots (Matplotlib/Seaborn) to fully interactive web visualizations
(Plotly/Altair).

o R (ggplot2, [Link] bindings): The gold standard for statistical visualization

(ggplot2) and often integrates with web visualization libraries.

4. Specialized Web Visualization Libraries

These libraries are used by front-end developers to create bespoke, highly interactive data
experiences embedded in web applications.

 Best for: Custom web applications, unique interactive visualizations, public-facing data
journalism projects.

 Top Choices:

o [Link] (Data-Driven Documents): The industry standard for bespoke web

visualizations, offering unparalleled control but requiring strong JavaScript
knowledge.

o Vega/Vega-Lite: Declarative languages for creating visualizations that are

simpler than [Link] and can export to various formats.

24 | P a g e
Decision Checklist

When making your choice, consider these factors:

1. Audience: Are you visualizing data for internal business users (BI tool), a developer
team (libraries), or the general public (web libraries)?

2. Skillset: Can your team code in Python or R, or do they prefer a drag-and-drop interface?

3. Data Source: Where does your data live? Ensure the software connects easily to your
databases, APIs, or files.

4. Interactivity: Do you need static charts for reports, or interactive dashboards with filters
and drill-downs?

25 | P a g e