💸 Loan Repayment Prediction

📘 Project Overview

This project focuses on predicting whether a borrower will fully repay their loan (loan_paid_back) using a combination of financial, demographic, and credit-related features. Loan repayment prediction is a critical task for financial institutions, as accurate risk assessment helps reduce defaults, optimize lending strategies, and ensure regulatory compliance.

The dataset contains nearly 600,000 records, capturing a diverse population of borrowers with a wide range of incomes, credit scores, employment statuses, and loan purposes. By this rich dataset, the project demonstrates real-world applications of machine learning in finance, including data preprocessing, feature engineering, model selection, hyperparameter tuning, evaluation, and deployment.

🔍 Key Highlights

• Comprehensive Data Analysis: Understanding the relationships between borrower characteristics, loan terms, and repayment behavior.
• Feature Engineering: Creating informative features such as debt-to-income ratios, credit score categories, and interaction terms to improve predictive performance.
• Multiple Modeling Approaches: Comparing traditional machine learning models (Random Forest, XGBoost, LightGBM) with neural networks to identify the best-performing approach.
• Explainable AI: Using feature importance analysis and SHAP values to understand why the model makes certain predictions, which is essential for trust and regulatory purposes.
• Interactive Deployment: Providing an end-user interface (via Streamlit) where users can input borrower information and receive a predicted repayment probability in real-time.

The goal is not only to build an accurate predictive model but also to create a reproducible and interpretable end-to-end machine learning pipeline.

By combining data-driven insights, model performance, and interpretability, this project highlights the practical application of machine learning in financial risk assessment and serves as a strong portfolio piece for data science professionals.

---

🧠 Dataset Information

Features:

Category	Columns
Financial	annual_income, debt_to_income_ratio, loan_amount, interest_rate
Credit	credit_score, grade_subgrade
Demographic	gender, marital_status, education_level, employment_status
Loan Info	loan_purpose
Target	loan_paid_back

• Rows: ~593,994
• Target variable: loan_paid_back (float; 1 = fully repaid, 0 = default or partial)
• Data Type: Mixed numerical and categorical variables

---

🧩 Project Workflow

1. Data Preprocessing

   • Handling missing values
   • Encoding categorical features(One-Hot Encoding, Ordinal Encoding)
   • Feature scaling and normalization
   • Outlier detection (IsolationForest)

2. Exploratory Data Analysis (EDA)

   • Visualizing distributions, correlations and mutual information
   • Understanding feature importance
   • Detecting skewness and transformations (Yeo-Johnson)

3. Feature Engineering

   • Creating new features such as income-to-loan ratio, risk index, etc.
   • Removing redundant or low-variance features

4. Model Development

   • Algorithms tested:
     • Logistic Regression
     • Random Forest
     • XGBoost / LightGBM
     • Neural Network (Keras / TensorFlow)
   • Hyperparameter tuning using Optuna and Keras Tuner

5. Model Evaluation

   • Metrics: AUC-ROC (Area Under the Receiver Operating Characteristic Curve)
   • Cross-validation (K-Fold)
   • Feature importance

---

🛠️ Technologies Used

Category	Libraries
Data Processing	pandas, numpy
Visualization	matplotlib, seaborn, plotly, streamlit
Machine Learning	scikit-learn, xgboost, lightgbm
Deep Learning	TensorFlow, Keras, PyTorch
Optimization	optuna, keras_tuner
Statistics	scipy, statsmodels

---

🚀 Future Work

• Deployment
  • Streamlit app for interactive prediction
  • Visualizing model metrics and feature influences dynamically
• Integrate external credit history datasets for improved accuracy
• Add time-series analysis for sequential loan performance prediction
• Experiment with ensemble stacking models
• Deploy the app on Streamlit Cloud or AWS EC2
• Incorporate explainable AI dashboards using Plotly Dash or Streamlit + SHAP

---

📈 Example Use Case

A bank analyst inputs the following borrower data into the Streamlit dashboard:

Feature	Value
annual_income	72,000
debt_to_income_ratio	0.35
credit_score	680
loan_amount	12,000
interest_rate	7.8
education_level	Bachelor
employment_status	Full-time

The trained model outputs:
👉 Predicted probability of full repayment: in progress

---