Credit Scoring Model Using Machine Learning

Project Overview

This project develops and compares machine learning classification models to predict credit risk using historical financial and applicant information.

The project uses the German Credit dataset from the UCI Machine Learning Repository.

Objective

To classify applicants into Good Credit Risk and Bad Credit Risk categories using supervised machine learning techniques.

Dataset Information

Dataset: Statlog (German Credit Data)

Source: https://archive.ics.uci.edu/dataset/144/statlog+german+credit+data

Property	Details
Total Records	1,000
Original Features	20
Good Credit Risk	700 (70%)
Bad Credit Risk	300 (30%)
Training Records	800
Testing Records	200

The dataset includes credit amount, credit duration, credit history, employment information, savings status, age, and other financial or demographic attributes.

Data Preprocessing

The following preprocessing techniques were applied:

• Missing value and duplicate record verification
• Exploratory Data Analysis (EDA)
• Feature engineering using credit amount per month
• One-Hot Encoding of categorical variables
• StandardScaler for numerical features
• Stratified train-test split (80:20)

The preprocessing process expanded the original 21 input features (including the engineered feature) into 62 numerical features.

Machine Learning Algorithms

Three classification models were implemented:

1. Logistic Regression
2. Decision Tree Classifier
3. Random Forest Classifier

Model Performance

Model	Accuracy	Precision	Recall	F1-Score	ROC-AUC
Logistic Regression	78.50%	67.35%	55.00%	60.55%	0.8057
Decision Tree	72.50%	57.14%	33.33%	42.11%	0.6859
Random Forest	76.50%	69.70%	38.33%	49.46%	0.7927

Precision, Recall, and F1-Score refer to the Bad Credit Risk class.

Best Performing Model

Logistic Regression achieved the best overall results among the three evaluated models.

• Accuracy: 78.50%
• ROC-AUC: 0.8057
• Bad Risk Recall: 55.00%

The trained model and preprocessing pipeline were saved together using Joblib.

Prediction Example

The saved model was tested on a held-out applicant record.

Model Prediction: Good Credit Risk

Estimated Bad Credit Risk Probability: 22.76%

Actual Class: Good Credit Risk

The model correctly classified this sample.

Technologies Used

• Python
• Pandas
• NumPy
• Matplotlib
• Seaborn
• Scikit-learn
• Joblib
• UCI ML Repository
• Google Colab

Installation

Clone this repository:

git clone https://github.com/Suchi0012/CodeAlpha_CreditScoringModel.git

Navigate to the project directory:

cd CodeAlpha_CreditScoringModel

Install the required dependencies:

pip install -r requirements.txt

How to Run

1. Download or clone the repository.
2. Open CodeAlpha_CreditScoringModel.ipynb in Google Colab or Jupyter Notebook.
3. Install the required Python libraries.
4. Execute the notebook cells sequentially.
5. Review the visualizations, model performance comparison, and predictions.

The dataset is retrieved using the ucimlrepo package.

Project Structure

CodeAlpha_CreditScoringModel/
├── CodeAlpha_CreditScoringModel.ipynb
├── credit_scoring_model.pkl
├── requirements.txt
├── README.md
└── .gitignore

Limitations

This project is intended for educational purposes only. The models have not undergone comprehensive fairness testing, probability calibration, or external validation.

The models should not be used to make real-world lending or credit approval decisions.

Conclusion

This project demonstrates the application of machine learning classification algorithms to credit risk prediction.

Logistic Regression achieved the strongest overall performance among the evaluated models, with 78.50% accuracy and an ROC-AUC of 0.8057.

Author

Sadia Afrin Suchi