🏦 Transparent Credit Scoring with Explainable Machine Learning

A machine learning project focused on building an interpretable and accurate credit scoring system, developed for the Engineering Foundations in FinTech (IEDA4500) course at HKUST.

🧠 Project Objectives

• Classify individuals into Good, Standard, or Poor credit categories
• Maintain high accuracy while improving transparency using SHAP explainability

⚙️ Techniques & Tools

• Languages & Libraries: Python, scikit-learn, XGBoost, SHAP, pandas, seaborn, matplotlib
• Feature Engineering:
  • Debt-to-Income (DTI)
  • EMI-to-Salary ratio
• Models Compared:
  • Logistic Regression
  • Random Forest (selected)
  • XGBoost

🔍 Explainability

• Used SHAP values to interpret both global and local model behavior
• Identified top drivers of credit risk: Monthly Balance, Payment Behaviour, Age

📊 Results

• Accuracy: ~80%
• Macro F1-Score: 0.79
• Generated CSV with credit score probabilities for downstream integration

📂 Files

• notebooks/: Model training and SHAP explanation scripts
• report.pdf: Final documentation
• slides.pdf: Presentation deck

🧰 Skills Demonstrated

Explainable AI (XAI) · Credit Risk Modeling · Model Evaluation · Feature Engineering · FinTech

🔗 Connect

LinkedIn • Email