Diabetes Classification (XGBoost)

Lightweight notebook that cleans the UCI-style diabetes survey data and trains an XGBoost classifier to predict the Diabetes_binary flag.

What the notebook does

• Load data from data/diabetes.csv and inspect missing values.
• Clean: median-fill the single BMI NaN, drop one row with missing HeartDiseaseorAttack, and mode-fill Fruits plus other sparse nulls.
• Feature prep:
  • Normalize BMI into bins and scale BMI, Physical_Health, Mental_Health, and Age to 0-1.
  • One-hot encode categorical columns: General_Health, Sex, Education, Income.
  • Split labels/features (Diabetes_binary target).
• Modeling: train/test split (80/20), baseline XGBClassifier with tuned learning rate, depth, estimators, subsample, and early stopping.
• Evaluation: accuracy, precision, recall, confusion matrix; optional GridSearchCV over learning rate, max depth, estimators, and column sampling.

How to run

1. Ensure Python 3.9+ and install deps:
   • pip install pandas numpy scikit-learn xgboost matplotlib.
2. Open and run the notebook classifier.ipynb top-to-bottom.
3. Confirm data/diabetes.csv exists relative to the notebook (place it under ./data/).

Key files

• classifier.ipynb: data cleaning, feature engineering, XGBoost training, evaluation, and hyperparameter search.
• data/diabetes.csv: input dataset (not committed; provide locally).

Notes

• Hyperparameter search uses ROC AUC via GridSearchCV; best params are reused to re-evaluate train/test scores.
• Confusion matrices are plotted for both train and test sets via plot_confusion_matrix.