Skip to content

README.md

Latest commit

 

History

History
190 lines (129 loc) · 4.35 KB

README.md

File metadata and controls

190 lines (129 loc) · 4.35 KB

💳 Credit Risk Prediction System

A Machine Learning project that predicts whether a loan applicant is Low Risk or High Risk using structured financial and demographic data.

🚀 Live Streamlit App: https://credit-risk-prediction-system-1.streamlit.app/

📊 At-a-Glance Impact

Metric Value
Dataset Size 1,000+ Credit Applications
Models Trained Decision Tree, Random Forest, Extra Trees, XGBoost
Best Accuracy 85%+ (Balanced Classes)
Target Variable Credit Risk (Good / Bad)
Business Goal Reduce Default Risk & Improve Loan Decisions

🚀 Why This Project Matters

Banks and financial institutions lose billions of dollars annually due to loan defaults.

🔴 Even a 1% improvement in credit risk prediction can save millions in bad loans.

This project simulates a real banking credit approval system, applying machine learning to:

  • Reduce default risk
  • Improve approval accuracy
  • Enable data‑driven lending decisions

📊 Dataset Overview

  • Customers: ~1,000 loan applicants
  • Features: Demographic + Financial attributes
  • Target: Credit Risk (Good / Bad)

Key Features Used

Feature Description
Age Applicant age
Sex Gender
Job Employment category
Housing Own / Rent / Free
Saving Accounts Savings status
Checking Account Checking balance
Credit Amount Loan amount requested
Duration Loan duration (months)
Purpose Loan purpose

🧠 Machine Learning Models

Multiple models were trained and evaluated:

Model Strength
Decision Tree Interpretability
Random Forest Stability & performance
Extra Trees Reduced variance
XGBoost High predictive power

⚡ XGBoost delivered the best overall performance.

📈 Results & Impact

Metric Score
Accuracy ~80%+
Recall (High Risk) Improved via class balancing
Precision Optimized using GridSearch

Business Interpretation

  • Correctly flags high‑risk borrowers
  • Minimizes false approvals
  • Improves lender profitability

🔧 Feature Engineering

  • Label Encoding for categorical variables
  • Target encoding for Risk variable
  • Class imbalance handled using class_weight="balanced"
  • Hyperparameter tuning with GridSearchCV

📦 Project Structure

credit-risk-prediction/
│
├── data/
│   └── german_credit.csv
│
├── models/
│   ├── decision_tree.pkl
│   ├── random_forest.pkl
│   └── label_encoders/
│
├── notebooks/
│   └── credit_risk_analysis.ipynb
│
├── src/
│   ├── preprocessing.py
│   ├── train.py
│   └── evaluate.py
│
├── README.md
└── requirements.txt

🛠 Tech Stack

  • Python 🐍
  • Pandas / NumPy
  • Scikit‑Learn
  • XGBoost
  • Seaborn & Matplotlib
  • Joblib

⚙️ How to Run

pip install -r requirements.txt
jupyter notebook

or train directly:

python src/train.py

📌 Key Learnings

  • Real‑world data is messy & imbalanced
  • Model accuracy alone is not enough — recall matters
  • Explainability is crucial in finance ML

🎯 Who Should Look at This?

✔ Recruiters (Data Science / ML / FinTech) ✔ Banks & NBFC analysts ✔ ML engineers learning applied finance ✔ Students preparing for placements

🌟 Future Improvements

  • SHAP explainability
  • Model deployment with Streamlit
  • Real‑time credit scoring API
  • ROC‑AUC optimization

🙌 Author

Prathmesh Bunde CSE | FinTech | Machine Learning

📌 If this project helped you, don’t forget to ⭐ the repo!