A/B-Testing-Project, E-commerce Conversion Optimization

📌 Overview

This project demonstrates how A/B testing can be applied to measure the impact of product or design changes on conversion rates. The dataset comes from an e-commerce experiment where users were randomly assigned to: - Control Group → old website design - Treatment Group → new website design The business goal is to evaluate whether the new design improves conversion rates.

🎯 Objectives

1. Define and test hypotheses for conversion improvement.
2. Explore and visualize the dataset.
3. Perform a two-proportion z-test to check statistical significance.
4. Compute confidence intervals for conversion rate differences.
5. Conduct power analysis to ensure the test is adequately powered.
6. Translate results into business insights & recommendations.

🗂️ Dataset

• Columns:
  • user_id → Unique identifier for users
  • timestamp → Date of visit
  • group → Control (old) or Treatment (new)
  • landing_page → Page version shown
  • converted → 1 if user converted, 0 otherwise
• Source: Kaggle – E-commerce A/B Testing Dataset

🔍 Methodology

1. Data Cleaning
   • Removed duplicates and missing values.
   • Checked balance between control & treatment groups.
2. Exploratory Data Analysis (EDA)
   • Conversion rates by group.
   • Distribution of users across groups.
   • Visual comparisons (bar plots, line trends).
3. Statistical Testing
   • Null Hypothesis (H₀): Conversion rates are equal.
   • Alternative Hypothesis (H₁): Conversion rates are different.
   • Applied two-proportion z-test with α = 0.05.
   • Calculated 95% confidence interval for conversion difference.
4. Power Analysis
   • Checked statistical power (>0.8 preferred).
   • Ensured sample size was adequate for detecting meaningful effects.

📊 Results

 - Z-statistic: 1.19
 - P-value: 0.2323
 - 95% Confidence Interval for difference: [-0.0009, 0.0038]

✅ Interpretation

• Since p-value > 0.05, we fail to reject the null hypothesis.
• The confidence interval for the difference includes 0, suggesting no statistically significant improvement in conversion.
• In other words, the new design did not significantly outperform the old design.

💡 Business Insights

• The test suggests rolling out the new design may not yield higher conversions.
• Further experiments could explore: - Segment-level testing (e.g., by device type, geography). - Alternative design changes or content personalization. - Running the test longer to collect more data.

🛠️ Tech Stack

• Python (pandas, numpy, matplotlib, seaborn)
• Statsmodels (proportions z-test, confidence intervals, power analysis)
• Google Collab for analysis and storytelling