Data Visualization: Charts, Dashboards & Insights

Why Visualize Data?

Data visualization converts complex datasets into intuitive, actionable insights. Humans process visuals 60,000x faster than text (MIT study). For example:

• eSewa uses heatmaps to detect fraudulent transaction clusters.
• NTC plots network traffic graphs to predict outages.
• Pathao employs real-time ride demand maps to optimize driver routes.

Core Principles of Effective Visualization

1. Clarity: Avoid clutter; focus on one key message.
2. Accuracy: Distortions (e.g., truncated axes) mislead.
3. Audience: Tailor to stakeholders (e.g., executives vs. engineers).
4. Storytelling: Guide the viewer’s eye with annotations and color.

mindmap
  root((Data Visualization Principles))
    Clarity
      Avoid: "Chartjunk" (decorative elements)
      Use: White space, minimal labels
    Accuracy
      Rule: "Lie Factor" = (Size of effect in graph)/(Size in data) ≤ 1.2
    Audience
      Example: Bar charts for comparisons, line charts for trends
    Storytelling
      Tools: Annotations, color gradients, guided flow

1. Types of Visualizations

A. Univariate (Single Variable)

Shows distribution of one variable.

• Histograms: Frequency of values (e.g., age distribution of Daraz customers).
• Box Plots: Median, quartiles, outliers (e.g., NEPSE stock price volatility).

Box plot showing median, quartiles, and outliers for NEPSE stock returns. (Image: Schlurcher, CC BY 4.0, via Wikimedia Commons)

• Pie Charts: Proportions (use sparingly—humans misread angles).

B. Bivariate (Two Variables)

Reveals relationships.

• Scatter Plots: Correlation (e.g., Pathao ride demand vs. weather).

• Line Charts: Trends over time (e.g., Ncell monthly revenue growth).

  # Example: Scatter plot of Daraz sales vs. marketing spend (Python)
  import matplotlib.pyplot as plt
  plt.scatter(x=sales, y=marketing_spend, color='blue')
  plt.xlabel("Marketing Spend (USD)")
  plt.ylabel("Sales Volume")
  plt.title("Daraz: Does More Spend = More Sales?")
  plt.show()
  

C. Multivariate (Three+ Variables)

Layered insights.

• Heatmaps: Intensity (e.g., eSewa fraud hotspots by district).
• Bubble Charts: Size + color (e.g., NTC’s internet speed vs. latency vs. user count).
• Treemaps: Hierarchical data (e.g., Khalti transaction categories).

Heatmap showing eSewa fraud rates across Nepal’s districts (red = high risk). (Image: Simonsarris, CC0, via Wikimedia Commons)

2. Choosing the Right Chart

3. Design Best Practices

A. Color Theory

• Categorical data: Distinct colors (e.g., Tableau’s default palette).
• Sequential data: Gradients (e.g., blue→red for low→high).
• Accessibility: Avoid red-green (colorblind users); use tools like ColorBrewer.

B. Axes and Labels

• X-axis: Independent variable (e.g., time, categories).
• Y-axis: Dependent variable (e.g., sales, temperature).
• Labels: Use plain language (e.g., "Monthly Revenue (USD)" not "Rev").

C. Annotations

• Highlight key data points (e.g., "Peak fraud in Kathmandu: 20%").
• Use arrows, text boxes, or tooltips (interactive dashboards).

4. Tools for Visualization

5. Real-World Example: Daraz’s Sales Dashboard

Problem: Daraz wants to identify which product categories drive revenue during festivals. Solution: A multipage dashboard with:

1. Line chart: Monthly sales trend (2023).
2. Bar chart: Top 5 categories by revenue (Tihar vs. Dashain).
3. Heatmap: Hourly sales peaks (e.g., 8–10 PM on weekends).
4. Scatter plot: Price vs. units sold (identifying price-sensitive items).

Outcome: Daraz increased festival sales by 18% by targeting high-demand categories with dynamic pricing.

6. Common Pitfalls & How to Avoid Them

7. Advanced Techniques

A. Geospatial Visualization

• Maps: Show location-based data (e.g., Pathao’s ride density in Kathmandu).

B. Interactive Visualizations

• Filters: Let users drill down (e.g., Tableau’s "Show Me" feature).
• Tooltips: Hover details (e.g., exact sales figures in Daraz’s dashboard).

C. Storytelling with Visuals

1. Hook: Start with a surprising stat (e.g., "Nepal’s e-commerce grew 40% in 2023").
2. Context: Explain the "why" (e.g., "Post-lockdown digital shift").
3. Insight: Highlight patterns (e.g., "Mobile payments rose 60% in rural areas").
4. Call to Action: Suggest next steps (e.g., "Invest in rural digital literacy").

In the Real World

1. eSewa

   • Idea Used: Anomaly detection heatmaps + time-series line charts.
   • How: Flags unusual transaction patterns (e.g., sudden large withdrawals) in real time. Engineers use Python’s Plotly to overlay fraud alerts on user activity maps.

2. NTC (Nepal Telecom)

   • Idea Used: Network traffic graphs (stacked area charts).
   • How: Tracks bandwidth usage by district to predict congestion. Example: During Dashain, Kathmandu’s data usage spikes by 300%—NTC pre-allocates bandwidth.

3. Pathao

   • Idea Used: Dynamic ride demand heatmaps + optimization algorithms.
   • How: Drivers see a live color-coded map (green = high demand) and reroute automatically. During festivals, surge pricing is triggered by scatter plots of ride requests vs. driver availability.

Exam Tip

1. Theory Questions:

   • Define lie factor, preattentive attributes (color, size), and small multiples.
   • Compare bar vs. line charts (use case, pros/cons).
   • Explain how colorblind-friendly palettes improve accessibility.

2. Practical Questions:

   • Given a dataset, choose the right visualization (e.g., "Show NEPSE stock trends" → line chart).
   • Critique a bad chart: Identify truncated axes, poor labels, or misleading colors.
   • Python/Tableau Task: Write code or describe steps to create a specific plot (e.g., "Plot Khalti transaction volumes by hour").

3. Case Studies:

   • Expect questions like:
     • "How would you visualize Daraz’s customer churn data?" → Cohort analysis line chart.
     • "Design a dashboard for NTC to monitor internet outages." → Geospatial map + alert system.

4. Tools:

   • Know the basics of matplotlib (Python) or Tableau’s drag-and-drop interface.
   • Example code snippet for a scatter plot (see above) may appear in the exam.

Key Formula to Remember: Lie Factor = $\frac{\text{Size of effect in graph}}{\text{Size of effect in data}}$ (Avoid charts where this > 1.2!)

Summary Checklist:

• Can you name 3 univariate and 3 multivariate charts?
• Do you know when to use a heatmap vs. a treemap?
• Can you critique a poorly designed chart?
• Have you practiced creating visuals in Python/Tableau?
• Can you tie visualization to a real Nepalese company’s use case?