Credit Risk Intelligence Platform

Predict defaults, quantify losses, and stress-test portfolios — all in one system.
Built on the UCI Taiwan Credit Card Default dataset.

🎯 What This Platform Demonstrates

Technical Approach

This system implements a champion/challenger model architecture for credit risk scoring. A WOE-based Logistic Regression scorecard (champion) runs alongside an XGBoost classifier (challenger), enabling real-time comparison of interpretable and ensemble-based risk signals.

  • ✓ 10-feature WOE scorecard via scorecardpy with monotonic score bands
  • ✓ Feature engineering: IV-based selection, multicollinearity removal, regulatory filtering
  • ✓ sklearn pipelines with StandardScaler / OrdinalEncoder per model
  • ✓ FastAPI backend with a business-friendly input layer mapping plain-English fields to raw features
What This System Does

This system combines machine learning and financial engineering to:

  • 📈 Predict Default Risk: Estimate probability of default (PD) for individual borrowers
  • 📊 Assign Credit Scores: Generate interpretable credit scores (576–906)
  • 📉 Analyze Portfolios: Segment borrowers into risk buckets
  • 💰 Calculate Expected Losses: Measure ECL = PD × LGD × EAD
  • 🎲 Run Simulations: Use Monte Carlo to estimate loss distributions
  • ⚠️ Stress Test: Evaluate risk under adverse economic scenarios

🔴 Problem & Solution

❌ Problem

Financial institutions face uncertainty in credit decisions:

  • No consistent methodology
  • Subjective judgment calls
  • Inability to assess portfolio-level risk
  • Poor visibility into tail risks
  • Regulatory non-compliance (ECL, IFRS 9)
✓ Solution

This system provides data-driven risk assessment:

  • ⚙️ Automated and consistent scoring
  • 📊 Interpretable credit scores & risk levels
  • 💡 Portfolio-level analytics and insights
  • 🎯 Quantified risk metrics (PD, LGD, ECL, VaR)
  • ✅ Regulatory-compliant methodology

🌟 Key Features

💡 Suggested Flow:
Start with Prediction → Analyze Portfolio → Run Simulation → Stress Test → Sensitivity Analysis

Six core capabilities for comprehensive credit risk management

🤖
Prediction Engine
Predict borrower default probability using two models: Logistic Regression (champion) and XGBoost (challenger)
📊
Risk Segmentation
Group borrowers into risk buckets for easier portfolio analysis and decision-making
💰
Expected Credit Loss
Calculate ECL (PD × LGD × EAD) to quantify expected portfolio losses. Essential for regulatory reporting
🎲
Monte Carlo Simulation
Run thousands of scenarios to estimate loss distributions. Calculate VaR and CVaR for tail risk analysis
⚠️
Stress Testing
Evaluate portfolio performance under Base, Mild, and Severe stress scenarios. Assess resilience to economic shocks
📈
Sensitivity Analysis
Identify key risk drivers. Measure how changes in PD and LGD impact total losses

🏗️ System Architecture

Frontend Layer: Modern, interactive HTML/CSS/JavaScript dashboard
API Layer: FastAPI backend with RESTful endpoints
ML Layer: Credit risk models (Logistic Regression + XGBoost)
Feature Layer: Engineered features, WOE transformations, binning
USER Browser / Dashboard HTTP API LAYER FastAPI Backend · REST Endpoints FEATURE LAYER Input Mapper · WOE Transform CHAMPION WOE Scorecard Logistic Regression · 10 features CHALLENGER XGBoost Classifier Gradient Boosting · Full feature set PD Score · Risk Band · ECL

Click any feature above to see this architecture in action

💡 Business Impact

📌 Data & Assumptions: This demo uses the UCI Taiwan Credit Card Default dataset (30,000 records) with assumed LGD and EAD values. All calculations follow standard credit risk formulas used in banking (IFRS 9 / Basel III).

🚀 Start Your Risk Analysis

Choose a feature below to begin your credit risk analysis

📬 Contact

Interested in this project or want to collaborate? Feel free to reach out.

📧 ompatel2587@gmail.com
📧 Send Email 🔗 LinkedIn Profile