Machine Learning for Diabetes Risk Prediction

xgb_model = XGBClassifier()
xgb_model.fit(X_train, y_train)
y_proba = xgb_model.predict_proba(X_test)[:, 1]
# test AUC: 0.993

why

a model sitting in a notebook doesn't help anyone. i wanted the prediction to be one form-submit away.

how it works

15,000 clinical records are cleaned down to 10,471, then random forest and XGBoost are compared; XGBoost won (0.993 vs 0.983 AUC) and ships behind a FastAPI endpoint.

what i learned

the feature-engineering pass moved the needle more than swapping algorithms did.

results

0.993 AUC and 96.1% accuracy on 2,095 held-out records.

the FastAPI endpoint returns a diabetes probability from the saved XGBoost model.