Skip to content

README.md

Latest commit

 

History

History
160 lines (103 loc) · 2.69 KB

README.md

File metadata and controls

160 lines (103 loc) · 2.69 KB
Raw

Loan Default Risk Prediction

Project Overview

This project builds a machine learning model to predict whether a customer is likely to default on a loan. It covers the complete ML pipeline from data preprocessing to model evaluation and hyperparameter tuning.

The objective is to help financial institutions identify high-risk customers and make better lending decisions.

Dataset

The dataset contains customer financial and demographic details such as:

  • Age, Income, Savings
  • Monthly Expenses
  • Credit Score
  • Loan Amount & Loan Term
  • Employment Years
  • Home Ownership
  • Education
  • Marital Status
  • Region
  • Recent Default History

Target Variable

target_default_risk

  • 0 → No Default
  • 1 → Default

Project Workflow

1. Data Cleaning

  • Handled missing values using median imputation
  • Fixed categorical inconsistencies (e.g., spelling issues)
  • Removed irrelevant columns like customer_id

2. Exploratory Data Analysis (EDA)

  • Value counts and countplots for categorical features
  • Distribution analysis of numerical features
  • Correlation heatmap
  • Outlier detection using boxplots

3. Feature Engineering

Created new meaningful features:

  • income_per_dependent
  • savings_to_income_ratio

These features help capture financial pressure and stability.

4. Data Preprocessing

  • Train-test split (80-20)
  • Ordinal Encoding for education
  • One-Hot Encoding for categorical variables
  • StandardScaler for numerical features
  • Outlier treatment using IQR-based capping (Winsorization)

5. Models Implemented

The following models were trained and evaluated:

  • Logistic Regression
  • Decision Tree
  • Support Vector Machine (SVM)
  • Random Forest
  • XGBoost

6. Hyperparameter Tuning

  • Used RandomizedSearchCV and GridSearchCV

  • Focused on improving:

    • Random Forest
    • XGBoost

7. Evaluation Metrics

Models were evaluated using:

  • Accuracy
  • Precision
  • Recall
  • F1 Score
  • Confusion Matrix

Results

  • Random Forest and XGBoost performed the best
  • XGBoost achieved the highest accuracy after tuning
  • Feature engineering significantly improved model performance

Technologies Used

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

How to Run

  1. Clone the repository
git clone https://github.com/prabandkumar/loan-default-risk-prediction.git
  1. Install dependencies
pip install -r requirements.txt
  1. Run the notebook
jupyter notebook

Future Improvements

  • Build a web app using Flask or FastAPI
  • Add model explainability (SHAP values)
  • Deploy model on cloud
  • Improve feature engineering