Predictive modeling · Deep learning · ETL & data engineering · SQL database design · Process improvement
Built a reproducible geospatial pipeline mapping pediatric specialist supply vs. child population across North Carolina's 100 counties. Identified 67% of counties as severe access deserts and 8 counties with zero pediatric specialists.
Five-source geospatial analysis of prenatal care access across NC's 100 counties; identified 27 counties with zero OB/GYN providers, 96/100 counties with active maternity shortage designations, and a federal surveillance gap that renders all 23 Western NC counties invisible in CDC natality data.
Personal fitness optimization using Six Sigma DMAIC, regression, and hypothesis testing; +200 cal/day improvement validated via two-sample t-test.
Normalized 3NF schema (8 entities) with T-SQL triggers and views for LARC (150+ members); migrated from Access → SQL Server for scalability and integrity.
ETL pipeline integrating 2 federal data sources across 2,956 counties; identified 326 "triple burden" counties where diabetes, uninsured rates, and poverty converge.
Heart-disease prediction on 319K+ CDC survey records; logistic regression with SMOTE achieved 83.5% ROC-AUC and 75.7% F1-score across 6 classifiers. Identified age, stroke history, and self-reported health as the dominant risk factors. Applied the trained pipeline to 445K unseen 2022 records.
Image classifier detecting melanoma from 10,015 dermoscopy images using two-stage transfer learning with EfficientNet-B0. Achieved Macro ROC-AUC of 0.925 across 7 lesion types. Grad-CAM heatmaps visualize model focus regions for clinical interpretability.