- Problem
- Independent pharmacies lack access to the drug-shortage prediction models large chains rely on, a methodological gap I confirmed across 5+ peer-reviewed and government sources (GAO, AJHP, PLOS).
- Approach
- Engineered a Python pipeline on the openFDA REST API to merge multiple datasets and surface early signals on shortage-risk factors. Validated the risk-scoring model with hypothesis testing and 5-fold cross-validation, benchmarked against published logistic-regression and ML-classifier studies (69–93% accuracy).
- My Role
- Sole researcher and engineer: sourced the literature, defined the methodology, built the data pipeline, and am now designing the pharmacy-facing interface.
- Outcome
- AUC 0.81
Cross-validated model performance landing within the published 69–93% accuracy benchmark range, now being built into a pharmacy-facing risk-scoring dashboard that turns FDA regulatory data into an actionable planning tool.
- Python
- openFDA API
- ML Classifiers
- Logistic Regression
