Loan Approval Predictor

An end-to-end Machine Learning project that predicts whether a loan will be approved or rejected based on applicant demographics, financial profile, and loan details. Built with a full ML pipeline including hyperparameter tuning, fairness analysis, and model explainability and deployed as a Streamlit web app.

---

Problem Statement

Predict a loan applicant's approval status given inputs like age, income, employment experience, credit score, loan amount, education level, and loan purpose.

Target variable: loan_status (binary classification problem: 0 = Rejected, 1 = Approved)

---

App Preview

---

Architecture

Raw Data → EDA → Outlier Removal → Data Preprocessing → SMOTE Balancing
                                                               ↓
User Input → Streamlit Web App → Voting Classifier Pipeline → Approved / Rejected
                                                               ↑
                                                        voting_classifier.pkl

---

Project Structure

Loan-Approval-Predictor/
│
├── .vscode/
├── artifacts/
│   ├── voting_classsifier.pkl
├── .gitignore
├── README.md
├── app.py
├── app_screenshot.png
├── loan-approval-predictor.ipynb
└── requirements.txt

---

Input Features

Feature	Type	Example
Age	Numeric	28
Annual Income	Numeric	55000
Employment Experience	Numeric	4
Credit Score	Numeric	700
Loan Amount	Numeric	12000
Interest Rate (%)	Numeric	10.5
Loan Percent Income	Numeric	0.22
Credit History Length	Numeric	6
Education	Categorical	Bachelor
Gender	Categorical	male / female
Home Ownership	Categorical	RENT / OWN / MORTGAGE / OTHER
Loan Purpose	Categorical	EDUCATION / MEDICAL / VENTURE / PERSONAL / HOMEIMPROVEMENT / DEBTCONSOLIDATION
Previous Loan Default	Categorical	Yes / No

---

Models Evaluated

Trained and compared using Optuna hyperparameter tuning, scored by AUC-ROC, with SMOTE applied to handle class imbalance:

• Logistic Regression
• Decision Tree
• Random Forest
• Gradient Boosting
• XGBoost
• LightGBM
• CatBoost
• AdaBoost
• Voting Classifier (Ensemble)
• Stacking Classifier

---

Results

Model	Accuracy	Precision	Recall	F1	AUC-ROC
Voting Classifier	0.9258	0.8884	0.8789	0.8835	0.9732
GradientBoosting	0.9259	0.8910	0.8752	0.8828	0.9730
LightGBM	0.9250	0.8901	0.8731	0.8812	0.9729
XGBoost	0.9244	0.8875	0.8746	0.8808	0.9729
Stacking Classifier	0.9220	0.8902	0.8610	0.8745	0.9684
RandomForest	0.9023	0.8405	0.8761	0.8563	0.9656

---

Fairness & Explainability

• Fairness Analysis performed using fairlearn measuring Demographic Parity Difference (DPD) and Equalized Odds Difference (EOD) across gender and education groups
• Bias Mitigation applied using ThresholdOptimizer to reduce DPD while preserving accuracy
• Model Explainability via SHAP summary plots, bar plots, and waterfall charts to interpret predictions

---

Quickstart

Local Setup

git clone https://github.com/Priyesh-DS-Code/loan-approval-predictor.git
cd loan-approval-predictor

python -m venv venv && source venv/bin/activate   # Windows: venv\Scripts\activate
pip install -r requirements.txt

streamlit run app.py                              # starts app on http://localhost:8501

---

Tech Stack

Python · Scikit-learn · XGBoost · LightGBM · CatBoost · Optuna · SMOTE · SHAP · Fairlearn · Streamlit

---

Connect

Priyesh · GitHub