Credit Risk Scoring Model using Machine Learning

Project Overview

This project focuses on building a Credit Risk Scoring System that predicts whether a loan applicant is likely to default or repay the loan.

The model is designed to help financial institutions make data-driven decisions by balancing:

• Risk (loan defaults)
• Profit (loan approvals)

---

📊 Dataset Information

• Source: Kaggle Credit Scoring Dataset

• Records: 261,384

• Features: 122

• Target:

• 0 → Non-Defaulter (Safe)

• 1 → Defaulter (Risky)

The dataset is imbalanced:

• 91.9% Non-default
• 8.1% Default

---

⚙️ Tech Stack

• Python
• Pandas, NumPy
• Scikit-learn
• Matplotlib, Seaborn
• Jupyter Notebook

---

Machine Learning Pipeline

🔹 Data Preprocessing

• Handled missing values using SimpleImputer
• Encoded categorical variables using OneHotEncoder
• Removed high-missing and irrelevant features
• Built clean preprocessing pipeline using ColumnTransformer

---

Feature Engineering

• Created:

  • Debt-to-Income Ratio
  • Payment Ratio
  • Employment Years

• These features help capture financial stability and repayment ability

---

Models Implemented

• Logistic Regression (baseline)
• Decision Tree Classifier
• Random Forest Classifier (Best Model)

---

Model Performance

Final Model: Random Forest Classifier

• ROC-AUC Score: ~0.715
• Threshold Tuned: 0.2

---

Final Confusion Matrix

[[45694  2361]
 [ 3360   862]]

---

📋 Classification Report

              precision    recall    f1-score
Non-Default     0.93       0.95       0.94
Default         0.27       0.20       0.23

---

⚖️ Threshold Optimization (Key Insight)

Different thresholds were tested:

Threshold	Behavior
0.1	High risk detection, many false alarms
0.2 ✅	Balanced (Final choice)
0.3	Conservative, misses defaulters

👉 Why important? In credit scoring:

• Missing a defaulter = high loss
• Rejecting a good customer = opportunity loss

---

Feature Importance (Top Factors)

• external_source_2
• external_source_3
• Age (days_birth)
• Employment history
• Debt-to-income ratio
• Loan amount & annuity

👉 External credit bureau scores were the strongest predictors

---

Business Impact

This model helps banks:

• ✅ Identify high-risk customers
• ✅ Reduce loan defaults
• ✅ Improve approval strategy
• ✅ Balance risk vs profit

---

🚀 How to Run

# Clone repository
git clone https://github.com/areebotix/Credit_Scoring_Model.git

# Navigate to project
cd Credit_Scoring_Model

# Install dependencies
pip install -r requirements.txt

# Run notebook
jupyter notebook

---

Project Structure

Credit_Scoring_Model/
│
├── data/
├── notebooks/
├── models/
├── src/
├── README.md
├── requirements.txt
└── .gitignore
``NOTE
SOME FILES ARE UNABLE TO UPLOAD DUE TO LARGE SIZE`

---

## 🎯 Key Learnings

* Handling **imbalanced datasets**
* Importance of **evaluation metrics beyond accuracy**
* Real-world trade-offs in **credit risk modeling**
* Power of **ensemble models (Random Forest)**
* Business impact of **threshold tuning**

## 📊 Results

### Confusion Matrix
![Confusion Matrix](outputs/confusion_matrix.png)

### ROC Curve
![ROC Curve](outputs/roc_curve.png)

---

## 👨‍💻 Author

**Areeba Khan**

---

## ⭐ If you like this project

Give it a star ⭐ on GitHub and feel free to connect!