MY WORK

Analytics Projects

Three end-to-end analyses. For each, I designed the research question, analytical framework, and variable selection logic, using AI tools to accelerate code implementation — the thinking is mine, documented and defensible.

🛒 01

E-Commerce Customer Retention Analysis

OLIST Brazilian Marketplace · 100K Orders

Pythonpandasstatsmodelsscikit-learn

The Question

What drives customers to come back — and which segments are worth retention spend?

The Approach

  • Cleaned and joined a 100K-order relational dataset across orders, payments, reviews, and geography
  • Built OLS and logistic regression models to identify repurchase drivers
  • Segmented the customer base with RFM (Recency-Frequency-Monetary) analysis
  • Translated model coefficients into concrete retention recommendations

Key Takeaway

Delivery experience and review scores dominate repurchase probability; a small high-value segment justifies targeted retention investment.

↗ View on Kaggle
📊 02

Mincer Wage Equation Analysis

CPS Microdata · ~64K Records

REconometricsBootstrapCross-validation

The Question

How much does an additional year of education actually raise wages — and does the classic Mincer model hold up?

The Approach

  • Estimated multiple specifications of the Mincer wage equation on ~64K CPS records
  • Ran five diagnostic tests covering heteroskedasticity, specification, and normality
  • Constructed bootstrap confidence intervals for key coefficients
  • Compared models via cross-validation rather than in-sample fit alone

Key Takeaway

Returns to education remain robust across specifications, but the experience profile deviates meaningfully from the textbook quadratic form.

↓ Download PDF
🎓 03

University ROI Analysis

Custom Financial Metrics · UCLA Deep-Dive

RFinancial ModelingData Visualization

The Question

Which universities deliver the best financial return on tuition — and how does UCLA compare?

The Approach

  • Designed custom ROI metrics combining cost, completion rates, and earnings outcomes
  • Ranked institutions on risk-adjusted return rather than raw earnings
  • Conducted a self-initiated deep-dive on UCLA outcomes by major
  • Visualized cost-vs-earnings tradeoffs for decision support

Key Takeaway

Sticker price is a poor predictor of value: several public institutions dominate elite privates once completion risk is priced in.

↓ Download PDF

Want the full methodology walkthrough for any of these?

Let's Talk ↗