Data & Analytics Projects

A selection of academic, professional, and personal projects spanning SQL, R, Python, and Tableau. Click any card to view the code, dashboard, or full write-up.

Olist E-Commerce Analysis

  • Analyzed ~100k orders across 9 relational tables in SQL (DuckDB) to identify what drives customer satisfaction and repeat purchases for a Brazilian e-commerce marketplace.
  • Found that late deliveries — just 6.7% of orders — cause ~37% of all 1-star reviews, and that most delays happen after the seller hands off the package.
  • Caught data issues that inflate common analyses of this dataset, including join fan-out and same-day multi-seller orders, revealing a true repeat rate of ~2.2% rather than the reported 3%.
  • Delivered five data-backed recommendations, including an early-warning system using missed seller handoffs, which predict late delivery at 5x the normal rate.
2026 — SQL (DuckDB), Python
Bar chart showing review scores dropping sharply once orders miss their estimated delivery date

CFB Power Ratings

  • Built a pipeline pulling multi-season game, roster, recruiting-talent, and SP+ data from the CollegeFootballData API into a documented Elo-style rating engine — an objective alternative to human-voted polls.
  • Backtested against every FBS-vs-FBS game since 2019 with no look-ahead: 71.4% accuracy, beating a home-favorite baseline by 12.6 points.
  • Automated the full pipeline with a scheduled GitHub Action that pulls new results and recomputes rankings every week.
  • Designed a live, dependency-free dashboard (vanilla JS + Chart.js) showing current rankings and each team's rating trajectory.
2026 — Python, GitHub Actions, Chart.js
CFB Power Ratings dashboard showing backtest accuracy results

Titanic Survival Prediction

  • Built an end-to-end classification pipeline predicting Titanic passenger survival, including feature engineering (Title extraction, Family Size, Cabin_Assigned flag) and VIF-based multicollinearity checks.
  • Trained and compared six models — Logistic Regression, Random Forest, XGBoost, Gradient Boosting, SVM, and KNN — using stratified 5-fold cross-validation.
  • Tuned Random Forest, XGBoost, and Gradient Boosting via GridSearchCV; evaluated all models with ROC/AUC analysis.
  • Selected tuned Random Forest as the final model — 83.7% validation accuracy, 0.890 AUC — and generated the Kaggle submission file.
2026 — Python, scikit-learn, XGBoost
ROC curves comparing six models on the Titanic dataset

MLB ROTO Score Prediction

  • Built a full machine learning pipeline to predict MLB player ROTO scores using the bat2022 dataset.
  • Performed exploratory data analysis, applied MLB batting qualification rules, explored positional performance patterns, and developed predictive models using Linear Regression and Ridge Regression.
  • Achieved an impressive R-squared of 0.987, validated through cross-validation.
Fall 2025 — Python, scikit-learn
Feature correlation heatmap from the MLB ROTO project

Predicting 2022 MLB ROTO Value

  • Modeled fantasy baseball (ROTO) value for hitters using 2022 MLB data and performance metrics such as SLG, AB, HR, H, BB, and SB.
  • Fit multiple linear regression models in R and compared goodness-of-fit using AIC, adjusted R², and residual diagnostics.
  • Explored interactions and multicollinearity, applying model selection techniques to find a parsimonious specification.
  • Built Tableau visualizations (scatterplots, residual plots, and player comparison views) to communicate model insights.
2025 — R, Tableau
Poster summarizing the 2022 MLB ROTO value project

Chocolate Bar Ratings: ANOVA & Effect of Cocoa %

  • Analyzed the "Flavors of Cacao" dataset to understand how company location and cocoa percentage impact expert ratings.
  • Used factorial ANOVA with Type II sums of squares to handle unbalanced data and interpret main effects and interactions.
  • Performed model diagnostics (residual plots, normality checks, variance homogeneity) and applied transformations where needed.
  • Summarized results in publication-style tables and plots using ggplot2 and reported practical implications for chocolate makers.
2025 — R, ANOVA
ANOVA results plot for the chocolate bar ratings project

CalEnviroScreen 4.0: Environmental & SocioEconomic Burden Analysis

  • Analyzed 8,035 census tracts across California using CalEnviroScreen 4.0 environmental and demographic indicators.
  • Investigated regional disparities by comparing average Pollution Burden Scores between Northern vs. Southern California using a two-sample t-test.
  • Identified statistically significant differences in pollution burden (p-value ≈ 1.92 × 10⁻²⁰¹) and created a county-level choropleth map to visualize geographic patterns.
  • Examined socioeconomic impacts by modeling the relationship between unemployment rate and pollution burden using simple linear regression.
2025 — R
Choropleth map of pollution burden across California counties

World Happiness Index: Regression & Random Forest Modeling

  • Analyzed the World Happiness Index & Inflation dataset (1,232 observations → 576 usable after cleaning), containing economic, demographic, and social indicators for 148 countries (2015–2023).
  • Explored global patterns using visualizations including histograms, scatterplots, and a full pairwise correlation matrix.
  • Built a Random Forest model achieving R² = 0.8117, outperforming linear regression and highlighting non-linear relationships between predictors and happiness.
  • Synthesized insights showing that economic strength and social well-being jointly drive national happiness levels, while high inflation and corruption negatively impact scores.
2025 — R
Correlation matrix from the World Happiness Index project