Credit Risk Analyzer

Project Description

A sophisticated credit risk model that categorizes loan applications into Poor, Average, Good, and Excellent categories based on criteria similar to the CIBIL scoring system.

def get_rating(score):
    if 300 <= score < 500:
        return 'Poor'
    elif 500 <= score < 650:
        return 'Average'
    elif 650 <= score < 750:
        return 'Good'
    elif 750 <= score <= 900:
        return 'Excellent'
    else:
        return 'Undefined'

Key Features

• Predict credit risk based on user inputs:
  • Age
  • Income
  • Loan Amount
  • Loan Tenure (Months)
  • Average DPD per Delinquency
  • Delinquency Ratio
  • Credit Utilization Ratio
  • Number of Open Accounts
  • Residence Type
  • Loan Purpose
  • Loan Type

Technical Details

• Programming Language: Python
• Libraries/Frameworks: scikit-learn, streamlit, pandas, numpy, joblib, xgboost

Installation

1. Clone the repository:

   git clone <repository-url>
   cd <repository-folder>

2. Install the dependencies:

   pip install -r requirements.txt

3. Run the Streamlit app:

   streamlit run app.py

Usage

1. Open the deployed app on Streamlit Cloud.
2. Enter user details such as:
   • Age
   • Income
   • Loan Amount
   • Loan Tenure (Months)
   • Average DPD per Delinquency
   • Delinquency Ratio
   • Credit Utilization Ratio
   • Number of Open Accounts
   • Residence Type
   • Loan Purpose
   • Loan Type
3. The app predicts:
   • Default Probability: e.g., 63.25%
   • Credit Score: e.g., 520
   • Rating: e.g., Average

Dataset Details

• Merged datasets:
  • Customers Dataset
  • Loans Dataset
  • Bureau Dataset
• Resulted in 50,000 rows and 33 columns.

Machine Learning Pipeline

1. Data Splitting: Performed train-test split early to avoid data leakage.
2. Data Cleaning: Replaced invalid loan purpose values with mode.
3. Feature Selection: Utilized VIF, IF, and domain knowledge.
4. Scaling: Min-Max scaling for numeric features.
5. Train-Test Split: 75% train, 25% test.
6. Model Training:
   • Logistic Regression
   • XGBoost
   • Random Forest
7. Fine Tuning:
   • RandomizedSearchCV
   • Optuna
8. Evaluation Metrics:
   • AUC-ROC
   • KS
   • Gini Coefficient
   • Classification Report

Deployment

• Platform: Streamlit Cloud

Screenshots

Feature Importance

License

This project is licensed under the Apache 2.0 License.

Contributions

Contributions are welcome! Feel free to open issues or submit pull requests.