Credit Classification with XGBoost

This project builds and evaluates a credit risk classification model using XGBoost and Scikit-Learn.
The goal is to predict whether a loan applicant is likely to be a good or bad credit risk, based on structured features from the classic German Credit dataset.

---

📋 Project Overview

Pipeline summary:

1. EDA: dataset inspection, class balance, and feature typing.
2. Feature Engineering: numeric / categorical separation, encoding, and data preparation.
3. Modeling: baseline XGBoost training with log-loss objective.
4. Evaluation: accuracy, ROC-AUC, confusion matrix, and class-wise metrics.
5. Hyperparameter Tuning: randomized search with stratified 5-fold CV.
6. Threshold Optimization: F1/precision-recall trade-off for class 0 (bad credit).
7. Model Export: final tuned model serialized with joblib.

---

📁 Repository Structure

credit-classification-xgboost/
├── data/                     # Input data (not versioned)
│   └── german_credit_cleaned.csv
├── notebooks/
│   └── 01_eda.ipynb          # Main Jupyter workflow
├── models/
│   └── xgb_best_model.joblib # Trained XGBoost model
├── src/
│   └── credit_risk/          # Future Python package (helpers, pipelines)
├── scripts/                  # Automation or CLI utilities
├── reports/                  # Figures, plots, analysis outputs
│   └── figures/
├── docs/                     # Optional documentation
├── requirements.txt          # Environment dependencies
└── README.md

---

⚙️ Environment

Python 3.13 (virtual environment)

Install dependencies:

pip install -r requirements.txt

---

🚀 Quick Start

Run the main notebook:

jupyter notebook notebooks/01_eda.ipynb

Outputs:

• Baseline + tuned XGBoost metrics
• ROC curve and threshold analysis
• Serialized model in models/xgb_best_model.joblib

---

📝 Notes

• StratifiedKFold ensures balanced folds for the binary target.
• Randomized search (40 iterations) achieved ROC-AUC ≈ 0.79.
• Decision threshold tuning improved interpretability for loan approval policy.
• The notebook is modular — each markdown section corresponds to a distinct training phase.

---