Bank Credit Scoring Model

Overview

A machine learning pipeline to predict loan default probability, with:

• A CLI script for full analysis and artifact generation
• A Streamlit dashboard for interactive EDA, model training, metrics, and live predictions

Project Structure

• credit_scoring.py — CLI pipeline: EDA, preprocessing, training, metrics, plots
• credit_scoring_app.py — Streamlit dashboard (upload CSV → explore → train → evaluate → predict)
• CreditScoring.ipynb — Original notebook (will be removed later)

Dataset

Typical columns:

• SeriousDlqin2yrs (target: 0/1)
• age, DebtRatio, MonthlyIncome, NumberOfDependents
• Payment delinquency counts, utilization metrics, etc.

Features

• EDA: distributions, correlation matrix
• Preprocessing: missing values, basic encoding for categoricals
• Modeling: Logistic Regression baseline
• Evaluation: confusion matrix, ROC AUC, PR curve, classification report
• Feature importance (coefficients)
• Streamlit app: upload data, interactively analyze and predict

Requirements

pip install -r requirements.txt

Main libraries: numpy, pandas, seaborn, matplotlib, scikit-learn, streamlit.

Dependencies and requirements.txt

Keep and commit requirements.txt for reproducible installs (local, CI, deployment).

• Create and activate a virtual environment:

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
python -m pip install --upgrade pip

• Install dependencies from the repo file:

pip install -r requirements.txt

• Update requirements.txt after changing packages:

pip freeze > requirements.txt

Quick Start

CLI (generates plots and metrics to files):

python credit_scoring.py

Streamlit app:

streamlit run credit_scoring_app.py

Using the Streamlit App

1. Upload a CSV containing SeriousDlqin2yrs and features.
2. Explore: target distribution, numeric distributions, correlation matrix.
3. Train: one-click Logistic Regression with stratified split.
4. Evaluate: confusion matrix, ROC AUC, classification report, top features.
5. Predict: input feature values and get default probability with a simple recommendation.

Results (CLI)

The script saves:

• confusion_matrix.png
• roc_curve.png
• precision_recall_curve.png
• feature_importance.png
• age_distribution.png, debt_ratio_distribution.png, numerical_distributions.png, correlation_matrix.png

License

MIT — see LICENSE.

Contact

Open an issue on GitHub for questions or suggestions.