🏦 Credit Scoring AI — Machine Learning Project

A production-ready, end-to-end Machine Learning system for credit risk assessment
Built as part of the CodeAlpha Machine Learning Internship

🚀 Live Demo • 📊 Dataset • 🛠️ Installation • 📖 Documentation

---

📌 Project Overview

This project implements a complete, professional Credit Scoring System that predicts whether a loan applicant is creditworthy or likely to default — using real financial and personal data.

It covers the full ML lifecycle:

• Real dataset ingestion from UCI ML Repository
• Data cleaning, EDA, and feature engineering
• Training & comparing 5 ML classification models
• Hyperparameter tuning with GridSearchCV
• SHAP-based model explainability
• A polished, interactive Streamlit web dashboard
• Downloadable prediction reports

---

✨ Features

Feature	Description
📊 Real Dataset	UCI German Credit Dataset — 1,000 real applicants, 20 features
🔬 Deep EDA	Correlation heatmaps, 3D scatter plots, risk trend analysis
⚙️ Feature Engineering	Debt ratios, age groups, credit tiers, encoding & scaling
🤖 5 ML Models	Logistic Regression, Decision Tree, Random Forest, GBM, XGBoost
📈 Full Evaluation	Accuracy, Precision, Recall, F1, ROC-AUC, Confusion Matrix
🎛️ Hyperparameter Tuning	GridSearchCV with 5-fold cross-validation
🔮 Live Prediction	Interactive form with risk gauge and confidence breakdown
🧠 Explainable AI	SHAP values + feature importance for every prediction
📥 Export Reports	Download prediction as CSV or JSON
🌙 Dark Mode UI	Modern, polished Streamlit dashboard

---

🛠️ Tech Stack

Backend:          Python 3.10+
ML Framework:     Scikit-learn, XGBoost
Data:             Pandas, NumPy, SciPy
Visualization:    Plotly, Seaborn, Matplotlib
Web App:          Streamlit
Explainability:   SHAP
Model Saving:     Joblib

---

📊 Dataset

German Credit Dataset — UCI Machine Learning Repository
Created by: Prof. Dr. Hans Hofmann, Universität Hamburg (1994)

Property	Value
Samples	1,000
Features	20 (financial + personal)
Target	1 = Creditworthy, 0 = Default
Class Split	70% Good / 30% Bad
Missing Values	None
Source	UCI ML Repository

Key Features:

• Checking account status
• Credit history
• Loan purpose & amount
• Savings account balance
• Employment duration
• Age, housing, job category
• Number of dependents

---

🚀 Installation

Prerequisites

• Python 3.10 or higher
• pip

Step 1: Clone the repository

git clone https://github.com/yourusername/CodeAlpha_CreditScoring.git
cd CodeAlpha_CreditScoring

Step 2: Create virtual environment (recommended)

python -m venv venv
# Windows:
venv\Scripts\activate
# macOS/Linux:
source venv/bin/activate

Step 3: Install dependencies

pip install -r requirements.txt

Step 4: Train the model

python train_model.py

This downloads the real dataset, trains 5 models, tunes the best one, and saves credit_model.pkl.

Step 5: Launch the app

streamlit run app.py

Open http://localhost:8501 in your browser.

---

📁 Project Structure

CodeAlpha_CreditScoring/
│
├── app.py                      # Main Streamlit app (entry point)
├── train_model.py              # Standalone training pipeline
├── credit_model.pkl            # Saved model bundle (after training)
├── requirements.txt            # Python dependencies
├── README.md                   # This file
│
├── utils/                      # Core ML modules
│   ├── __init__.py
│   ├── data_loader.py          # Dataset loading, cleaning, feature engineering
│   ├── model_trainer.py        # Training, evaluation, tuning, prediction
│   └── visualizations.py      # All Plotly chart functions
│
├── pages/                      # Streamlit page modules
│   ├── __init__.py
│   ├── home.py                 # Landing page
│   ├── dataset_insights.py     # Data preview & statistics
│   ├── eda.py                  # Exploratory data analysis
│   ├── model_training.py       # Training UI & metrics
│   ├── prediction.py           # Live prediction form
│   ├── explainability.py       # SHAP & feature importance
│   └── about.py                # Project info & deployment
│
├── data/                       # Dataset files
│   ├── german.data             # Raw UCI dataset (auto-downloaded)
│   └── german_credit_cleaned.csv
│
├── models/                     # Saved model files
│   ├── random_forest_bundle.pkl
│   └── model_metadata.json
│
├── notebooks/                  # Jupyter notebooks (EDA & experiments)
│   └── credit_scoring_eda.ipynb
│
├── screenshots/                # App screenshots for README
└── assets/                     # Static assets

---

📈 Model Results

Model	Accuracy	Precision	Recall	F1 Score	ROC-AUC
Logistic Regression	~75%	~0.83	~0.84	~0.84	~0.78
Decision Tree	~72%	~0.80	~0.82	~0.81	~0.72
Random Forest	~79%	~0.86	~0.87	~0.87	~0.83
Gradient Boosting	~78%	~0.85	~0.86	~0.86	~0.82
XGBoost	~80%	~0.87	~0.87	~0.87	~0.84

Results vary slightly based on random seed and tuning. XGBoost or Random Forest typically performs best.

---

🖥️ Screenshots

Add screenshots to the screenshots/ folder and link them here after running the app.

screenshots/
├── home_page.png
├── eda_charts.png
├── model_comparison.png
├── live_prediction.png
└── shap_explanation.png

---

🚀 Deployment

Streamlit Cloud (Recommended — Free)

1. Push this repo to GitHub
2. Go to share.streamlit.io
3. Connect your repo, set app.py as the entry point
4. Deploy!

Render

# Start command:
streamlit run app.py --server.port $PORT --server.address 0.0.0.0

HuggingFace Spaces

• Choose Streamlit SDK when creating a Space
• Upload project files or connect GitHub
• Add sdk: streamlit to README YAML header

---

🔮 Future Improvements

• Deep learning model (PyTorch/Keras) comparison
• SMOTE oversampling for class imbalance
• FastAPI REST endpoint for programmatic access
• Batch prediction (upload CSV of applicants)
• Real-time drift monitoring
• Docker containerization
• CI/CD with GitHub Actions

---

📄 License

This project is open source and available under the MIT License.

---

👨‍💻 Author

CodeAlpha Internship Project
Machine Learning Track | Credit Scoring Task

---

🙏 Acknowledgements

• UCI ML Repository for the German Credit Dataset
• Prof. Dr. Hans Hofmann for creating the benchmark dataset
• CodeAlpha for the internship opportunity
• Scikit-learn, XGBoost, SHAP, Plotly, Streamlit open source communities

---

⭐ Star this repo if you found it helpful!
CodeAlpha ML Internship