End-to-end analysis of the CDC Behavioral Risk Factor Surveillance System (BRFSS) 2023 microdata to build interpretable risk models and actionable insights.

• Large-scale survey preprocessing (types, missing codes, codeframes).
• Class-imbalance handling (e.g., SMOTE/SMOTENC) and multicollinearity control.
• Model suite: Logistic Regression, Random Forest, XGBoost, CatBoost (+ Optuna tuning).
• Transparent evaluation (ROC-AUC, PR-AUC, F1/Recall @ tuned threshold, calibration).
• Explainability with SHAP for feature importance and per-prediction rationale.

BRFSS/
├─ BRFSS2023.ipynb   # EDA → preprocessing → modeling → interpretation
└─ README.md

python -m venv .venv && source .venv/bin/activate
pip install -U pip wheel
pip install pandas numpy pyarrow scipy scikit-learn imbalanced-learn shap             matplotlib plotly xgboost catboost optuna jupyter

Download 2023 BRFSS microdata and documentation from CDC. Keep raw files outside the repo if possible and set a path in the notebook:

DATA_DIR = "/path/to/brfss2023"

jupyter lab   # open BRFSS2023.ipynb and Run All

• Target/outcome: CHCOCNC1 (self-reported cancer diagnosis; Yes=1, No=0)
• Best model: Logistic Regression (slightly higher AUPRC; equal ROC-AUC to XGB)
• Test ROC-AUC / PR-AUC: 0.799 / 0.217 (LogReg) — 0.799 / 0.215 (XGB)
• Operating threshold & rationale: Threshold chosen to achieve ~0.75 recall on validation.
  • LogReg thr ≈ 0.0936 → Precision 0.184, Recall 0.741
  • XGB thr ≈ 0.0844 → Precision 0.180, Recall 0.773
• Base prevalence (weighted, test): ~7.9% → AUPRC lift ≈ 2.75× (LogReg), 2.74× (XGB)
• Top drivers (model-based): _AGE_G (older age bands), _DRDXAR2 (arthritis diagnosis status), _RACEPRV (race category); also signal from _SEX and _CHLDCNT.

• Survey raking weights retained for reporting; model training may be unweighted—report population metrics with weights where relevant.
• Self-reported measures and telephone survey design introduce known biases; use results as decision support, not causal claims.

• Weighted training and weighted metrics.
• Threshold tuning vs. cost matrix by stakeholder scenario.
• Package to FastAPI for batch/online scoring + SHAP dashboards.

Analysis code for research/education. Verify CDC terms before redistributing data.