Data Analytics Portfolio — Selected Notebooks

This repository contains two internship projects designed as concise, resume-ready case studies. Together they demonstrate practical data cleaning, exploratory analysis, statistical inference, simulation, and experiment decision-making.

Project 1 — Google Play Store Exploratory Data Analysis

• Objective: Analyze Google Play Store app data to identify category-level trends, engagement patterns, and actionable product insights.
• Dataset: Public Google Play Store app metadata (10,841 rows) including app name, category, rating, reviews, size, installs, type, price, content rating, and last updated fields.
• Data Preparation:
  • Handled missing values (Rating imputation with median; selective row removal for sparse missing fields).
  • Cleaned and converted key columns (Reviews, Size, Installs, Price) into numeric formats.
  • Engineered time features (Day, Month, Year) from Last Updated.
  • Removed duplicate apps to ensure cleaner category comparisons.
• Analysis Performed:
  • Univariate analysis for numeric and categorical features.
  • Skewness, kurtosis, and IQR-based outlier assessment.
  • Category popularity analysis (share of apps and top categories).
  • Install concentration analysis by category and content rating.
  • Correlation heatmap across numeric variables.
  • Bivariate and multivariate views: rating by app type, installs by category and type, and log-transformed installs/reviews for skewed distributions.
• Key Findings:
  • A small set of categories dominates both app volume and installs.
  • Reviews and Installs are strongly related.
  • Free apps generally drive higher installation reach.
  • Content rating and app type provide meaningful context when interpreting performance.
• Notebook: Project_1_Google_Playstore_EDA.ipynb

Project 2 — A/B Testing Analysis (Conversion Optimization)

• Objective: Evaluate whether a new checkout funnel (Variant B) improves conversion compared with the baseline (Variant A).
• Experiment Setup:
  • Simulated traffic for two variants with equal allocation.
  • Baseline assumptions: 10,000 visitors per variant, with conversion probabilities near 10% (A) and 12% (B).
  • Used binomial simulation to generate purchases (binary outcomes).
• Statistical Workflow:
  • Computed conversion rates for both variants.
  • Built 95% confidence intervals for each proportion using normal approximation.
  • Ran a one-sided two-proportion z-test (H0: pB <= pA, H1: pB > pA).
  • Interpreted z-statistic and p-value for decision-making.
• Sequential Monitoring Simulation:
  • Simulated batch-wise monitoring (100 users per batch, 60 batches).
  • Tracked cumulative lift and p-values over time.
  • Demonstrated how significance can be monitored during experiment rollout.
• Key Results:
  • Simulated conversion rates showed B outperforming A (about 11.34% vs 9.73%).
  • Confidence intervals were non-overlapping in the sampled run.
  • Hypothesis test returned a statistically significant result (p-value < 0.05).
  • Sequential simulation also detected significant positive lift for Variant B.
• Additional Concepts Covered:
  • Interpretation of binomial distribution, p-value, z-test, and confidence intervals.
  • Monte Carlo simulation concept for uncertainty, win probability, and robustness checks.
• Notebook: Project_2_ABtesting.ipynb

Technical Stack

• Environment: Python 3.8+ with Jupyter Notebook
• Libraries: pandas, numpy, matplotlib, seaborn, scipy, statsmodels

How To Run

1. Open notebooks in Jupyter Notebook or JupyterLab.
2. Install dependencies:

• A requirements.txt file is included at the repository root listing the core dependencies.

pip install -r requirements.txt

Contact

• Name: Dev Patel
• Email: [email protected]
• GitHub: https://github.com/devpatel0005