Data mining knowledge representation: Visualization techniques.

In data mining, knowledge representation refers to how the discovered

patterns, associations, or insights are presented to users in a way that is
comprehensible and actionable. Since mined knowledge is often complex
(rules, clusters, decision trees, networks, etc.),
visualization techniques are widely used to represent it.

1. What is Visualization?
 Visualization is the process of representing data graphically so that
patterns, trends, and insights become easier to understand.
 In data mining / data analysis, visualization is especially important
because datasets are often large, complex, and multi-dimensional.

2. Motivation for Visualization

Why do we use visualization in data mining and analytics?
 Humans process visual information faster than raw numbers.
 To discover hidden patterns and relationships in large datasets.
 To make comparisons (e.g., sales across regions).
 To communicate results effectively to non-technical users.
 To deal with high-dimensional data, where numerical inspection is
impossible.

3. General Concepts of Visualization Techniques

Visualization techniques can be classified based on:
1. Data Type:
o One-dimensional (histogram, line chart)
o Two-dimensional (scatter plot, heatmap)
o Multi-dimensional (parallel coordinates, radar charts, projections)

2. Representation Style:
o Geometric (scatter plots, projections)
o Icon-based (glyphs, Chernoff faces)
o Pixel-oriented (each value mapped to a pixel, used for large data)
o Hierarchical (tree maps, dendrograms)
o Graph-based (network visualization)
o
 3. Techniques of Data Visualization
Common Visualization Techniques
These techniques can be broadly categorized into a few main types:
1. Charts and Graphs:
Foundational tools for comparing data, showing trends, and
understanding distributions.
 Bar Charts: Ideal for comparing quantities across different

categories.
 Line Charts: Excellent for illustrating trends and changes over
time.
 Pie Charts: Used to show the proportion of a whole.
 Scatter Plots: Help in analyzing the relationship between two
variables.
 Histograms: Display the distribution of a dataset.
 Box Plots: Visualize data distribution, identifying medians,
quartiles, and outliers.

2. Maps:
Used to represent geographical data and highlight spatial
relationships.
 Heatmaps: Visualize the density of data points in a geographical
area or across categories.
 Choropleth Maps: Use colors to represent statistical data over
geographical
areas.
 3. Hierarchical and Part-to-Whole Visualizations:
Useful for data with nested or layered structures.
 Tree Maps: Represent hierarchical data using nested rectangles,
with the size of each rectangle proportional to its value.
 Sunburst Charts: Show hierarchical data in a radial layout.

4. Text-Based and Analytical Visualizations:

For understanding textual data and complex relationships.
 Word Clouds: Visually represent textual data by showing the
frequency of words, where larger words appear more frequently.
 Network Diagrams: Illustrate connections and relationships
between different entities.

Decision Trees: Visualize the decision-making process by representing a

set of rules.

Example: Retail
Store Sales Data
Analysis

Imagine you are analyzing a retail store chain’s sales dataset.

The dataset has attributes like:
  Date (daily/weekly)
 Region (North, South, East, West)
 Product Category (Electronics, Clothing, Groceries, etc.)
 Units Sold
 Revenue
 Customer Age Group

1️Line Chart → Revenue trend over months

2️Bar Chart → Total revenue by product category
3️Scatter Plot → Relationship between units sold and revenue
4️Heatmap → Revenue comparison across regions and categories

Line Chart( Revenue trend over months)

Bar Chart (Total revenue by product category)

 Scatter Plot→ Relationship between units sold and revenue

Heatmap→ Revenue comparison across regions and categories

Categories of Techniques Based on Approach

 These techniques are categorized by how they process and display
data:

 Data visualization techniques can be broadly classified into five main

categories, depending on the type and complexity of data to be
represented:
 These techniques are categorized by how they process and display
data:

 1. Pixel-Oriented Visualization Techniques: : Map each data value to

a single pixel, using color to represent the value.

 Idea: Each data value is represented as a single colored pixel.

 Use case: Very large datasets (millions of records).
 Examples:
o Recursive Pattern Technique
o Circle Segments Technique
Practical Techniques :Heatmaps, pixel maps
 Advantage: Can display a huge amount of data in a compact space.
 Limitation: Interpretation may be difficult without interaction.

 2. Geometric Projection Techniques: Use mathematical formulas to

project high-dimensional data into a lower-dimensional space, such
as scatter plot matrices.

 Idea: Map multi-dimensional data into 2D or 3D geometric spaces.

 Use case: To explore relationships and clusters in high-dimensional data.
 Examples:
o Scatter plots
o Principal Component Analysis (PCA) plots
o t-SNE and MDS visualizations
 Advantage: Helps to find clusters, correlations, and patterns.
 Limitation: May lose information during dimensionality reduction.

 3. Icon-Based Visualization Techniques: Use icons to represent multi-

dimensional data values.

 Idea: Represent each data object as a small icon/glyph, where different

attributes are encoded as visual features.
 Use case: Multivariate data analysis.
  Examples:
o Chernoff Faces (human face features mapped to variables)
o Star Plots / Radial Graphs
 Advantage: Human brain can easily detect similarities/differences in
icons.
 Limitation: Becomes confusing with too many variables or large
datasets.

4. Hierarchical Visualization Techniques :Organize data into a tree

structure, with nodes representing data elements and branches showing
their relationships.
 Idea: Used for data with hierarchical or tree-like structures.
 Use case: Taxonomies, organizational structures, decision processes.
 Examples:
o Tree Maps(Decision trees)
o Dendrograms
o Cone Trees
 Advantage: Good for showing nested relationships.
 Limitation: Hard to read when hierarchy is very deep.

5. Graph-Based Visualization Techniques

 Idea: Represent data objects as nodes and their relationships as edges.
 Use case: Social networks, web link analysis, knowledge graphs.
 Examples:
o Force-directed Graphs
o Association Graphs
 Advantage: Clearly shows relationships and connectivity.
 Limitation: May become cluttered for very dense graphs.

4. Visualizing Higher-Dimensional Data

Real-world datasets often have more than 3 attributes (dimensions), but our
eyes can only perceive 2D/3D.
Techniques to visualize high-dimensional data include:
 Parallel Coordinates: Each attribute is an axis; lines represent data
points.
 Dimensionality Reduction:
o PCA (Principal Component Analysis)
o t-SNE (t-distributed stochastic neighbor embedding)
 o UMAP (Uniform Manifold Approximation and Projection)
 Star/Radar Plots: Each axis is an attribute; data represented as polygons.
 Glyphs & Icons: Represent attributes with shapes, colors, or faces.
 Pixel-Oriented Techniques: Map values to pixel colors across the
screen.

5. Do’s and Don’ts of Visualization

✅ Do’s:
 Choose the right visualization for the data type and audience.
 Keep it simple and clear, avoid unnecessary decoration.
 Use consistent color schemes (avoid confusion).
 Label axes, legends, and units properly.
 Highlight important insights (not everything equally).
 Use interactive visualization if the dataset is huge.
❌ Don’ts:
 Don’t overload with too much information in one chart (avoid clutter).
 Don’t use misleading scales (e.g., truncated axis that exaggerates results).
 Don’t rely only on color (consider color-blind viewers).
 Don’t use 3D plots unnecessarily (often confusing unless needed).
 Don’t ignore context (data without explanation is meaningless).