🏦 Loan Approval Prediction using XGBoost + SHAP + KMeans clustering

This project is a comprehensive machine learning pipeline developed in Python to predict loan approval status based on applicant data. It includes meticulous preprocessing, robust modeling using XGBoost, hyperparameter optimization via 4-vector strategies, and post-hoc interpretability using SHAP to explain individual predictions. Also applying KMeans clustering, identifying natural groupings of loan applicants.

---

🚀 Project Objectives

• Predict whether a loan will be Approved or Rejected
• Build a robust XGBoost model with hyperparameter optimization
• Explain the predictions using SHAP visualizations and custom textual summaries
• KMeans clustering, Helped in identifying natural groupings of loan applicants
• Provide actionable feedback for rejected applications

---

📚 Dataset

• Format: CSV (; delimited)

• Dataset shape : Records : 45000, Columns : 14

• Credit_History

  • Target: Loan_Status (Y/N)

---

🧱 Workflow Overview

1. 📊 Data Cleaning & Exploration

• Dropped duplicate rows
• Visualized distributions using pie and bar charts
• Confirmed datatype consistency and field coverage

2. 🧼 Feature Engineering

• Encoded categorical variables using LabelEncoder
• Removed uninformative features (Home_Ownership, Loan_Purpose, etc.)
• Split dataset: 70% train / 15% validation / 15% test

3. 🧠 Model Building: XGBoost

• Baseline model trained using default XGBoost

4. 🧬 Optimization via 4-Vector Search

A novel 4-dimensional vector technique was used to search the hyperparameter space. This method iteratively tested combinations of four continuous variables to identify the most performant XGBoost setup based on validation accuracy.

5. 🔍 Explainability with SHAP

• SHAP TreeExplainer used to compute feature contributions

• Visuals:

  • Summary plots
  • Force plots for individual predictions

• Custom explanation DataFrame:

  • Loan_ID, outcome, top contributing reasons, and suggested advice
  • Example:
    • Rejected due to low Credit_History and high LoanAmount
    • Advice: "Try applying for a smaller loan or improving your credit history"

6. 🔵 Customer Segmentation with KMeans

• Performed KMeans clustering on numeric features (e.g., ApplicantIncome, LoanAmount)
• Assigned cluster labels as a new feature to enhance predictive power
• Helped in identifying natural groupings of loan applicants

---

📈 Visualizations

• 📊 Stacked bar plots for gender-wise approval rates
• 🥧 Pie chart of approval/rejection proportions
• 🧠 Interactive SHAP plots (summary, dependence, force)

---

🛠 Requirements

pip install pandas numpy scikit-learn xgboost shap matplotlib seaborn

🧪 Optional (Development & Notebook Use)
pip install jupyter notebook