💳 Credit Risk Modelling

A Machine Learning project to predict the probability of loan default using statistical techniques and advanced ML models. This project focuses on risk quantification, feature engineering, and model interpretability, similar to real-world banking systems.

🚀 Project Overview

Credit risk modelling is a critical task in financial institutions to assess whether a borrower is likely to default.

In this project, we:

Cleaned and validated raw loan applicant data Engineered domain-specific financial features Performed statistical testing and feature selection Built and evaluated classification models to predict default risk

🧠 Key Features

📊 EDA & Statistical Analysis

Chi-Square Test (Categorical features) Distribution analysis & correlation checks Data imbalance understanding

⚙️ Feature Engineering

Loan-to-Income Ratio Delinquency Metrics Debt-to-Income (DTI) Credit history-based features

📉 Feature Selection

Weight of Evidence (WOE) Information Value (IV) Variance Inflation Factor (VIF) for multicollinearity

🤖 Models Used

Logistic Regression (Baseline, interpretable)

XGBoost (Advanced boosting model)

📏 Evaluation Metrics

ROC-AUC

🏗️ Tech Stack

Languages: Python Libraries: Pandas, NumPy Scikit-learn XGBoost Matplotlib, Seaborn

📂 Project Structure

🔬 Model Evaluation Metric Purpose,

ROC-AUC Model discrimination ability,

📈 Key Insights

Feature engineering significantly improved model performance,

WOE-IV helped in selecting highly predictive variables,

XGBoost outperformed Logistic Regression in accuracy,

Logistic Regression provided better interpretability for risk scoring.

💡 Business Impact

Helps banks reduce default risk,

Enables data-driven loan approval decisions,

Improves credit scoring systems,

Supports regulatory-compliant risk modelling.