Credit Risk Modeling Pipeline

Overview

This repository contains a Python pipeline for credit risk prediction using machine learning. The project is based on the Home Credit Default Risk dataset and focuses on assessing the likelihood of a loan applicant defaulting. The pipeline includes data preprocessing, feature engineering, model training, evaluation, and result visualization.

Features

• Data Preprocessing: Handles missing values, normalizes numerical data, and encodes categorical and ordinal attributes.
• Feature Engineering: Selects relevant features for credit risk modeling.
• Machine Learning Models: Implements Logistic Regression, Random Forest, and XGBoost.
• Hyperparameter Tuning: Uses GridSearchCV to optimize the best model.
• Visualization: Generates confusion matrices and prediction distribution plots.
• Model Persistence: Saves trained models and predictions for further analysis.

Dataset

• Source: Kaggle - Home Credit Default Risk
• Target Variable: TARGET (1 = Default, 0 = No Default)

Key Features

Numerical Features:

• AMT_INCOME_TOTAL: Total income of the applicant.
• AMT_CREDIT: Total credit amount of the loan.
• DAYS_BIRTH: Age of the applicant in days (negative values).

Categorical Features:

• NAME_INCOME_TYPE: Type of income source (e.g., working, pensioner).
• OCCUPATION_TYPE: Type of occupation.

Ordinal Features:

• NAME_EDUCATION_TYPE: Level of education (e.g., secondary, higher).
• REGION_RATING_CLIENT_W_CITY: Rating of the region where the client lives.

Installation

Prerequisites

Ensure you have Python installed (>=3.7) and install the required dependencies:

pip install pandas numpy scikit-learn xgboost matplotlib seaborn joblib

Usage

• Clone the repository:

git clone https://github.com/your-username/credit-risk-modeling.git cd credit-risk-modeling

• Run the script:

python credit_risk_pipeline.py

• Outputs:

1. Classification Reports (*_classification_report.txt)

2. Confusion Matrices (*_confusion_matrix.png)

3. Prediction Results (*_predictions.csv)

4. Trained Models (*_model.pkl)

Results

• Accuracy & Performance Metrics: The model performance is evaluated using accuracy, precision, recall, and F1-score.

• Prediction Visualization: Histogram plots show the distribution of predicted defaults vs. actual outcomes.

Future Improvements

• Implement feature selection for better model performance.

• Explore additional models (e.g., Neural Networks, Gradient Boosting).

• Integrate real-world credit risk data.

License

This project is licensed under the MIT License.

Author:

Xikun Jiang For inquiries, contact: xikunjiang@163.com