Credit Risk Modeling Project

Overview

This project implements a credit risk modeling solution using logistic regression to predict loan default risk. It utilizes customer, loan, and credit bureau data to build a predictive model, incorporating data preprocessing, feature engineering, model training, and evaluation.

Project Structure

• Notebook: credit_risk_model_codebasics.ipynb - Main Jupyter Notebook containing the code for data loading, preprocessing, modeling, and evaluation.
• Dataset:
  • dataset/customers.csv: Contains customer demographic information (e.g., age, gender, income).
  • dataset/loans.csv: Contains loan details (e.g., loan amount, tenure, default status).
  • dataset/bureau_data.csv: Contains credit bureau data (e.g., open accounts, credit utilization).
• Artifacts:
  • artifacts/model_data.joblib: Saved model file containing the trained logistic regression model, feature names, scaler, and columns to scale.

Prerequisites

To run this project, ensure you have the following Python libraries installed:

• pandas
• numpy
• matplotlib
• seaborn
• scikit-learn
• joblib

You can install them using pip:

pip install pandas numpy matplotlib seaborn scikit-learn joblib

Data Description

The dataset consists of three CSV files:

1. customers.csv: Includes customer details such as:
   • cust_id: Unique customer identifier
   • age, gender, marital_status, employment_status, income, etc.
2. loans.csv: Includes loan details such as:
   • loan_id, cust_id, loan_purpose, loan_type, sanction_amount, default, etc.
3. bureau_data.csv: Includes credit bureau data such as:
   • cust_id, number_of_open_accounts, credit_utilization_ratio, delinquent_months, etc.

Each dataset contains 50,000 records, which are merged on cust_id for analysis.

Methodology

1. Data Loading and Merging:
   • Load the three datasets using pandas.
   • Merge customers.csv and loans.csv on cust_id, then merge with bureau_data.csv to create a unified dataset.
2. Feature Engineering:
   • Create new features like loan_to_income and avg_dpd_per_delinquency.
   • Encode categorical variables (e.g., residence_type, loan_purpose, loan_type) using one-hot encoding.
   • Scale numerical features using a scaler (saved in model_data.joblib).
3. Model Training:
   • Use logistic regression to predict the default column (True/False).
   • Split data into training and testing sets using train_test_split.
   • Train the model and evaluate feature importance based on model coefficients.
4. Model Saving:
   • Save the trained model, feature names, scaler, and columns to scale in artifacts/model_data.joblib.

Usage

1. Clone the repository or download the project files.
2. Ensure the dataset files are in the dataset/ directory.
3. Open and run the credit_risk_model_codebasics.ipynb notebook in a Jupyter environment.
4. The notebook will:
   • Load and preprocess the data.
   • Train the logistic regression model.
   • Display feature importance using a bar plot.
   • Save the model to artifacts/model_data.joblib.

Results

• The logistic regression model is trained to predict loan defaults.
• Feature importance is visualized to show which features (e.g., credit_utilization_ratio, loan_to_income) most influence the prediction.
• The model and preprocessing components are saved for future use.

Future Improvements

• Experiment with other algorithms (e.g., Random Forest, XGBoost) for better performance.
• Perform hyperparameter tuning to optimize the logistic regression model.
• Add cross-validation to ensure robust model evaluation.
• Include additional feature engineering to capture more complex patterns.