Credit Risk Modelling with LLM Integration

This enhanced version of your credit risk model integrates Large Language Models (LLMs) to provide intelligent, interpretable insights alongside your machine learning predictions.

🚀 Features

Original Features

• Credit risk prediction using logistic regression
• Credit score calculation (300-900 range)
• Risk rating (Poor, Average, Good, Excellent)
• Interactive Streamlit interface

New LLM-Enhanced Features

• AI-Powered Risk Analysis: Detailed explanations of risk factors
• Personalized Recommendations: Actionable advice for both lenders and borrowers
• Risk Factor Identification: Highlights key factors affecting creditworthiness
• Alternative Solutions: Suggests loan restructuring or mitigation strategies
• Natural Language Insights: Human-readable explanations of model outputs

📋 Setup Instructions

1. Install Required Packages

pip install streamlit openai joblib numpy pandas scikit-learn

Or create a requirements.txt:

streamlit==1.31.0
joblib==1.3.2
numpy==1.24.3
pandas==2.0.3
scikit-learn==1.3.0
openai

Then install:

pip install -r requirements.txt

2. Get Your openai API Key

1. Visit https://platform.openai.com/settings/organization/api-keys
2. Sign up or log in
3. Navigate to API Keys section
4. Create a new API key
5. Copy the key (starts with sk-proj-)

3. Set Up Environment Variable

Linux/Mac:

export openai_api_key='your-api-key-here'

Windows (Command Prompt):

set  openai_api_key=your-api-key-here

Windows (PowerShell):

$env: openai_api_key='your-api-key-here'

Or create a .env file:

 openai_api_key=your-api-key-here

Then load it in your code:

from dotenv import load_dotenv
load_dotenv()

4. File Structure

credit_risk_model/app
│
├── main.py                    # Enhanced Streamlit app
├── prediction_llm_helper.py       # ML + LLM integration                          # Original Streamlit app
├── artifacts/
│   └── model_data.joblib              # Your trained model

5. Run the Application

streamlit run main_with_llm.py

🎯 How It Works

1. Traditional ML Pipeline (Unchanged)

• Takes user inputs (age, income, loan details, etc.)
• Preprocesses data with scaling
• Uses logistic regression for prediction
• Calculates credit score and rating

2. LLM Enhancement Layer (New)

• Receives ML model outputs + input features
• Constructs a detailed prompt with financial context
• Calls Claude API for intelligent analysis
• Returns structured insights with:
  • Risk summary
  • Key risk factors
  • Lender recommendations
  • Borrower improvement tips
  • Alternative strategies

📊 Example Output

Traditional Output:

• Default Probability: 15.32%
• Credit Score: 678
• Rating: Good

Enhanced LLM Output:

### Risk Summary
This applicant presents a MODERATE risk profile with a Good credit rating...

### Key Risk Factors
1. **Positive**: Low loan-to-income ratio (2.13) indicates strong repayment capacity
2. **Concern**: Delinquency ratio of 30% suggests past payment issues
3. **Concern**: Average DPD of 20 days shows pattern of late payments

### Recommendations

**For the Lender:**
- Approve with conditions: higher interest rate (prime + 2-3%)
- Require additional collateral or guarantor
- Set up automated payment reminders

**For the Borrower:**
- Focus on consistent on-time payments for next 6-12 months
- Reduce credit utilization below 20%
- Consider debt consolidation to simplify payments

### Alternative Actions
- Offer a smaller loan amount initially (₹18L instead of ₹25.6L)
- Implement graduated payment structure
- Provide financial literacy counseling

🔧 Customization Options

1. Use Different LLM Providers

OpenAI (GPT-4):

from openai import OpenAI

client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))

response = client.chat.completions.create(
    model="gpt-4-turbo-preview",
    messages=[{"role": "user", "content": prompt}]
)
insights = response.choices[0].message.content

Google (Gemini):

import google.generativeai as genai

genai.configure(api_key=os.environ.get("GOOGLE_API_KEY"))
model = genai.GenerativeModel('gemini-pro')
response = model.generate_content(prompt)
insights = response.text

2. Customize the Prompt

Modify the prompt in get_llm_insights() to:

• Focus on specific risk factors
• Change the tone (more technical/casual)
• Add regulatory compliance checks
• Include industry-specific guidelines

3. Add Caching

For repeated requests with similar parameters:

import functools
from functools import lru_cache

@lru_cache(maxsize=100)
def get_llm_insights_cached(credit_score, rating, probability):
    # Cached version for similar scores
    pass

4. Streaming Responses

For real-time output in Streamlit:

with client.messages.stream(
    model="claude-sonnet-4-20250514",
    max_tokens=1500,
    messages=[{"role": "user", "content": prompt}]
) as stream:
    for text in stream.text_stream:
        st.write(text, end="")