All Projects

Finding: Zone trips got faster after the fee (+0.79 mph Uber, +0.41 mph Lyft), but a placebo test showed the higher Uber fares were an existing trend, not the fee.

Analyzed 3.55M Uber and Lyft trips with SQL and a difference-in-differences design, plus a placebo test, to measure what NYC congestion pricing changed for riders and drivers.

Finding: The rule-based system caught only 16 of 8,213 fraud transactions — a 99.81% miss rate concealing over $12B. Louvain recovered ~65% unsupervised (2,660 fraud account pairs).

Modeled 6.3M mobile money transactions across 9M+ accounts as a 15.4M-node Neo4j graph to detect fraud patterns invisible at the individual-transaction level.

Finding: Queens holds 36.89% of street trees versus Manhattan’s 9.36%, and tree coverage does not track neighborhood income.

Analyzed the 2015 NYC Street Tree Census with 2024 neighborhood income data in Tableau for the New York Restoration Project to surface environmental equity gaps in urban tree coverage.

Finding: Migrant students and English language learners face over 3× higher dropout rates; NYC districts average 13.2% versus 7.5% elsewhere, with enrollment size the strongest driver.

Analyzed 39,674 NYSED school district records (140 variables) to identify the factors most associated with student dropout across New York State.

Finding: Combining collaborative and content-based signals in one graph solves sparsity and cold-start where either approach alone fails.

Built a hybrid Neo4j movie recommender modeling users, movies, genres, and directors, combining Jaccard collaborative filtering with FastRP embeddings and Precision@K evaluation.

Finding: Logistic Regression achieved the highest test AUC-ROC (0.6959), outperforming XGBoost, Random Forest, MLP, and a soft-voting ensemble on Framingham CHD risk prediction.

Predicted 10-year CHD risk on the Framingham teaching dataset with KNN imputation, SMOTE balancing, clinical interaction features, and a seven-model bake-off including ensembles.