🌗 Light vs Dark Theme Website Performance

Analyzing User Behavior Insights & A/B Testing Website Metrics with Python

This project analyzes website performance metrics to compare "Light Theme" and "Dark Theme" using a dataset (data/abt.csv) with 1000 user records. Built in Python within a Jupyter Notebook (abt.ipynb), it leverages pandas, numpy, scipy, seaborn, matplotlib, and scikit-learn for data processing, statistical testing, and visualizations.

---

Dataset Overview

The dataset (data/abt.csv) contains website user interaction data with 1000 rows (486 Light Theme, 514 Dark Theme). A preview of key columns:

Theme	Click_Through_Rate	Conversion_Rate	Bounce_Rate	Scroll_Depth	Age	Location	Session_Duration	Purchases	Added_to_Cart
Light Theme	0.032	0.208	0.754	73.02	52	Chennai	1176	No	Yes
Light Theme	0.143	0.028	0.306	35.02	33	Pune	780	Yes	Yes
Dark Theme	0.323	0.179	0.297	45.74	27	New Delhi	912	No	No
Light Theme	0.489	0.326	0.649	79.37	41	Chennai	1345	No	Yes
Light Theme	0.098	0.273	0.438	72.18	36	Pune	645	Yes	Yes

• Size: 1000 rows (sample: 5 shown).
• Key Metrics: CTR, Conversion, Bounce, Scroll, Purchases, Demographics.

---

Analysis Sections and Visualizations

1. Data Summary

Provides descriptive statistics for numerical and categorical variables.

• Metrics: Means, std, ranges for Click_Through_Rate, Conversion_Rate, etc.
• Visualizations: None (tabular output).
• Insights:
  • Mean CTR: 0.256, Conversion: 0.253, Bounce: 0.506, Scroll: 50.32.
  • Age range: 18-65, Session: ~15.5min avg.

---

2. Theme Performance Comparison

Compares mean metrics between themes.

• Metrics: Click_Through_Rate, Conversion_Rate, Bounce_Rate, Scroll_Depth.
• Visualizations: Styled table.
• Insights:
  • Dark Theme CTR: 0.265 vs. Light 0.247 (higher).
  • Light Theme Conversion: 0.255 vs. Dark 0.251 (slight edge).

---

3. Hypothesis Testing

Tests for significant differences using Welch’s t-tests (alpha = 0.05).

• Metrics: Click_Through_Rate, Conversion_Rate, Bounce_Rate, Scroll_Depth.
• Hypotheses:
  • H₀: No difference in means.
  • H₁: Difference exists.
• Visualizations: Styled table with p-values.
• Insights:
  • CTR: p = 0.048 (significant, Dark > Light).
  • Conversion: p = 0.635 (not significant).

---

4. Correlation Analysis

Examines linear relationships between metrics.

• Metrics: Click_Through_Rate, Conversion_Rate, Bounce_Rate, Scroll_Depth.
• Visualizations: Heatmap (seaborn).
• Insights:
  • Weak correlations (< 0.3) → metrics are independent.

---

5. Distribution Analysis

Explores metric distributions by theme.

• Metrics: Conversion_Rate.
• Visualizations: Histogram with KDE (seaborn).
• Insights:
  • Light Theme: Bimodal (~0.15, ~0.40) → two user groups.
  • Dark Theme: Unimodal (~0.25) → consistent, lower conversions.

---

6. Location-Based Analysis

Assesses bounce rates by location and theme.

• Metrics: Bounce_Rate.
• Visualizations: Box plot (seaborn).
• Insights:
  • New Delhi: Dark Theme higher median (~0.60) vs. Light (~0.50).
  • Location > Theme influence.

---

7. Predictive Modeling

Predicts Purchases using logistic regression.

• Features: Click_Through_Rate, Conversion_Rate, Bounce_Rate, Scroll_Depth, Age.
• Visualizations: None (tabular coefficients).
• Insights:
  • Accuracy: ~60% (weak).
  • Positive: Conversion_Rate (0.02), Scroll_Depth (0.01).
  • Negative: Bounce_Rate (-0.03), Click_Through_Rate (-0.01).

---

Key Findings

• Theme Performance: Dark Theme boosts CTR significantly (p = 0.048); Light Theme slightly better for conversions (not significant).
• User Behavior: Weak metric correlations; bimodal Conversion in Light Theme suggests segmentation.
• Location Impact: New Delhi favors Light Theme (lower bounce).
• Prediction: Current features poorly predict purchases—needs non-linear models.

---

Methodology

• Tools: Python with pandas, numpy, scipy.stats, seaborn, matplotlib, scikit-learn.
• Visualizations: Static plots (heatmap, histogram, box plot) with seaborn.
• Processing: T-tests, correlation, regression, data cleaning (pd.to_numeric).

---

Usage

1. Setup: Install dependencies:

   pip install pandas numpy scipy seaborn matplotlib scikit-learn IPython
   

---

Author

Author: [Pavan Yellathakota]
Date: MAR 2025

---

Contact Information

You can reach out to me through the following channels:

• Email: pavanyellathakota@gmail.com
• LinkedIn: Pavan Yellathakota

For more projects and resources, check out:

• GitHub: Pavan Yellathakota
• Portfolio: pye.pages.dev

---

"""