Loan Defaulters Prediction System

Overview

This project presents a Loan Defaulter Prediction System developed to assess the likelihood of a customer defaulting on a loan, based on historical banking data. Using state-of-the-art machine learning techniques, the system predicts loan risk with a high degree of accuracy and is deployed using Flask for real-time predictions.

Objective

To build a robust machine learning model that predicts whether a customer will default on a loan using multiple business and financial indicators, and to deploy the model via a web interface.

Dataset Summary

• Source: Internal banking dataset
• Observations: 150,000 rows
• Features: 26 input variables including loan term, employment numbers, disbursement amount, franchise status, urban/rural category, SBA approval values, etc.
• Target: MIS_Status – Indicates whether the customer is a defaulter (CHGOFF) or non-defaulter (PIF)

Tools & Technologies Used

• Languages: Python
• Libraries: NumPy, Pandas, Matplotlib, Seaborn, Scikit-learn, XGBoost, LightGBM, Flask, Pickle
• Web Framework: Flask
• Model Deployment: HTML + Flask API
• Data Visualization: Matplotlib, Seaborn
• Model Evaluation: Accuracy, Confusion Matrix, ROC-AUC

Key Features

• Extensive data cleaning, imputation, and feature engineering
• Feature selection using ExtraTreesClassifier and XGBoost
• Model comparison using:
  • XGBoost (Accuracy: ~99%)
  • Random Forest
  • Naive Bayes
  • LightGBM
• Deployed web app allows real-time loan default prediction with a clean UI

Project Structure

Loan-Defaulters-Prediction/
├── app.py                    # Main Flask backend for deployment
├── Loan_final.py            # Model training and evaluation script
├── Model.pkl                # Serialized trained model
├── index.html               # Frontend user interface
├── style.css                # CSS styling for the UI
├── Requirement Project document.docx  # Project requirements documentation
├── Data dictionary.docx     # Description of all dataset variables
└── README.md                # Project documentation

Deployment

The web application is hosted using Flask with an HTML interface. The model was serialized using pickle and loaded into the Flask app for prediction.

How to Run

1. Clone this repo
2. Install dependencies using pip install -r requirements.txt
3. Run python app.py
4. Access the app via localhost:5000

Usage

The user inputs loan-related details such as:

• Loan Term
• Number of Employees
• Urban/Rural Indicator
• SBA Approval amount, etc.

The application predicts whether the loan is likely to default or safe to approve.

Impact

This system empowers banks and financial institutions to:

• Reduce loan default risk
• Speed up approval workflows
• Maintain financial health through predictive insights

Author

Surya Vamsi Patiballa
MS in Data Science at George Washington University

• Email :- svamsi2002@gmail.com
• LinkedIn :- https://www.linkedin.com/in/surya-patiballa-b724851aa/
• Resume :- https://drive.google.com/file/d/178IYcArC6YYVdJiIwRmJYodzKZ-JXe-D/view?usp=sharing

“Smart lending starts with smart predictions.”