15 moves ahead….
It means he SEES patterns. He sees: Now imagine you're a chess engine. You
- Where the pieces WANT to go can calculate 200 million
- What the opponent is threatening positions per second. But your output
- The FLOW of the position is: Score = 2.34.
MODULE 2: Data Visualisation & Exploration
WITH visualization: Patterns emerge.
Insights crystallize. Decisions are made.
some are good and
others are
garbage...
-Akanksha Rathore
3. CONFIRMATION
Use cases / Functions "My hypothesis is correct"
→ Control charts,
before-after plots, regression
residuals
1. EXPLORATION → Goal: Validate or refute a
"What am I looking at?" claim
→ Scatter plots, histograms,
distributions
→ Goal: Discover patterns YOU didn't
know existed
2. COMMUNICATION
"Here's what I found"
→ Bar charts, line graphs, time
series
→ Goal: Tell a story clearly to an
audience
Same stats! But all different patterns!
Y Y Y Y
│ ●● │ ● │ ● │ ●
│ ● ● │ ●● │ ●● │ ●●●●●●●
│ ● ● │ ●● │ ● ● │
│ ● ● │ ●●● │ ● ● │
│● ● │ ●● │● ● │
├─────────────→│←─────────────→│←────────────→│←──────────→
X X X X
Dataset 1 Dataset 2 Dataset 3 Dataset 4
Raw Data
↓
Clean/Preprocess
↓
VISUALIZE ← START HERE! (Not modeling!)
↓
Identify patterns/anomalies/questions
↓
Form hypothesis
↓
Model/Test
↓
Visualize results
↓
Communicate findings (with visuals)
↓
Insight
Key Visualization Types for Data Mining
Type 1: SCATTER PLOT (Relationships)
See if two continuous variables relate
Data structure: (X, Y) pairs
➔ Each point = one observation (multivariate thinking)
➔ Pattern emerges naturally (no summary hiding variation)
➔ Outliers visible immediately
Key Visualization Types for Data Mining
Type 2: HISTOGRAM (Distribution)
See the SHAPE of a single continuous variable
Data structure: One variable, many observations
✓ Shows variation (not hiding it in one number)
✓ Reveals outliers
✓ Shows skewness, bimodality
✓ No summary statistics can replace this
Key Visualization Types for Data Mining
Type 3: BOX PLOT (Comparison + Distribution)
Compare distributions ACROSS categories
Data structure: Multiple groups, continuous variable
✓ Compact (shows 5 numbers in one plot)
✓ Enables comparison (multiple boxes side-by-side)
✓ Outliers visible (marked separately)
✓ Not a summary (shows quartiles, not average)
Key Visualization Types for Data Mining
Type 4: HEATMAP (Multivariate Grid)
Show relationships across MANY variables (matrix data)
Data structure: M × N table with
numerical values
✓ Shows all cells (no averaging)
✓ Color encodes value (intuitive)
✓ Pattern = correlation structure becomes VISUAL
✓ Multivariate thinking (many variables shown)
Key Visualization Types for Data Mining
Type 5: LINE GRAPH (Time Series)
Show trends over TIME
Data structure: Time-indexed observations
✓ Time ordering is MEANINGFUL (connects dots, not
scatter)
✓ Trends visible (up, down, seasonal?)
✓ Change points obvious
✓ Simple and effective
Customer ID | Age | Income | Debt | Loan_Approved
1 | 25 | 35000 | 5000 | Yes
2 | 45 | 120000 | 2000 | Yes
3 | 32 | 48000 | 15000| No
... (1000 rows)
QUESTION: What visualization would YOU create?
To find predictors of approval?
To understand distributions?
To find variable relationships?
To present to boss?
Best practices
THE 6 TUFTE PRINCIPLES
PRINCIPLE 1: GRAPHICAL INTEGRITY
Definition: The representation of numbers on the graph must be
DIRECTLY PROPORTIONAL to the numbers they represent.
Real-world 2025 case:
COVID-19 death reporting - Some countries scaled axes
differently
to make their pandemic look better. GRAPHICAL
DISHONESTY.
THE 6 TUFTE PRINCIPLES
PRINCIPLE 2: DATA-INK RATIO
Tufte: "Maximize data-ink ratio. Eliminate decoration."
"Every ink mark should earn its place by telling data."
If you can remove something and lose NO INFORMATION,
REMOVE IT.
THE 6 TUFTE PRINCIPLES
PRINCIPLE 3: CHARTJUNK
Decorative or distracting visual elements that add
NO INFORMATION and REDUCE clarity.
Be ruthless about decoration.
If it doesn't serve data, it serves
DISTRACTION.
THE 6 TUFTE PRINCIPLES
PRINCIPLE 4: SMALL MULTIPLES
Repeat the same small graph many
times with different
data to enable easy comparison.
When to use: Comparing many similar
patterns across categories/time
THE 6 TUFTE PRINCIPLES
PRINCIPLE 5: MULTIVARIATE THINKING
Show MANY variables simultaneously,
not artificial summaries.
Context: Real-world problems need
multiple variables.
THE 6 TUFTE PRINCIPLES
PRINCIPLE 6: APPROPRIATE CHARTTYPE
RIGHT CHOICE = Matches data structure
Time series as pie chart (WHY??)
Continuous distribution as bar chart (loses
shape)
Geographic data as scatter plot (loses
location)
[Link]
tion-mistakes-and-how-to-avoid-them-e3896fe8e104
[Link]
hat-you-can-learn-from/