dworsleytonks/credit-scoring-xgb

Credit Scoring XGBoost

Summary

A baseline XGBoost model for binary credit risk classification trained on the Home Credit dataset. The model is intended for experimentation and educational use, not production.

Model Details

• Model type: XGBoost classifier
• Task: Binary classification (default risk)
• Input: Tabular features engineered from Home Credit raw tables
• Output: Probability of default
• Artifact: credit_scoring_xgb.pkl

Intended Use

• Demonstrate a credit scoring pipeline with MLflow tracking
• Provide a lightweight baseline for experimentation

Limitations

• Not validated for real-world credit decisions
• No fairness or regulatory compliance audit
• Performance depends on data preprocessing used in the training pipeline

How to Use

Load the artifact locally with joblib:

import joblib
model = joblib.load("credit_scoring_xgb.pkl")
preds = model.predict_proba(X)[:, 1]

Input Expectations

• The model expects the same feature engineering used in the training pipeline.
• See the project notebooks for the exact preprocessing steps and column names.

Reproducibility

• Training code and experiments are tracked in the main project repo:
  • src/train.py
  • notebooks/01_data_preparation.ipynb
  • notebooks/03_model_comparison.ipynb

Training Data

Home Credit (public dataset). CSVs are tracked via Git LFS in the project repo.

Metrics

See the project repo notebooks for evaluation details and MLflow logs.