Implemented predictive machine learning pipelines in Python using scikit-learn (Random Forest, Gradient Boosting), achieving a ROC-AUC of 0.871 and 87.1% accuracy in classifying urban food desert status across 800+ census tracts.
- Addressed severe class imbalances in demographic datasets by implementing SMOTE (imbalanced-learn), increasing the predictive recall of the at-risk minority class by 129% (from 31% to 71%) for downstream ML models.
- Designed and integrated geospatial data pipelines using GeoPandas to ingest, clean, and fuse 5+ disparate municipal datasets to enable longitudinal analysis of SNAP participation from 2015 to 2025.
- Developed an interactive, data-driven web dashboard utilizing D3.js to dynamically simulate variable income thresholds ($20k–$130k) and visualize real-time geospatial impacts.