💳 Loan Default Predictor

A machine learning web app that predicts whether a loan applicant is likely to default on their loan. Compares four classification models using ROC-AUC and 5-fold cross-validation, and shows which financial factors drive default risk through feature importance analysis.

Demo

Enter applicant details → AI evaluates credit risk → Instant prediction with default probability and Low / Medium / High risk rating.

Results

Model	Accuracy	ROC-AUC	CV AUC
Logistic Regression	~85%	~93%	~92%
Random Forest	~99%	~99%	~99%
Gradient Boosting	~99%	~99%	~99%
XGBoost ⭐	~99%	~99%	~99%

Features

• 4 models compared: Logistic Regression, Random Forest, Gradient Boosting, XGBoost
• ROC-AUC evaluation — the right metric for imbalanced credit risk data
• 5-fold cross-validation to verify model generalises beyond training data
• Feature importance chart showing exactly what drives default risk
• Interactive prediction form — enter any applicant's details and get a risk score
• Dataset auto-downloads on first run — no manual setup needed

Tech Stack

Layer	Technology
ML Models	Scikit-Learn, XGBoost
Preprocessing	StandardScaler
Evaluation	ROC-AUC, 5-Fold Cross-Validation
Web Framework	Flask
Dataset	Credit Risk / Lending Data
Language	Python

How to Run

1. Clone the repo

git clone https://github.com/manny2341/loan-default-predictor.git
cd loan-default-predictor

2. Install dependencies

pip install -r requirements.txt

3. Start the app

python3 app.py

The dataset downloads automatically on first run.

4. Open in browser

http://127.0.0.1:5015

How It Works

1. Applicant financial data entered via the prediction form
2. All features scaled with StandardScaler
3. XGBoost model predicts the probability of default
4. Risk level classified as Low (<35%), Medium (35–60%), or High (>60%)
5. Feature importance from Random Forest shows what matters most (e.g. debt-to-income ratio, interest rate)

Why ROC-AUC Over Accuracy?

Credit risk datasets are imbalanced — most applicants don't default. A model that always predicts "no default" still scores high accuracy but is useless. ROC-AUC measures how well the model separates defaulters from non-defaulters across all thresholds, making it the correct metric for this problem.

Project Structure

loan-default-predictor/
├── app.py               # Flask server, model training, feature importance, prediction API
├── dataset/
│   └── loan.csv         # Auto-downloaded on first run
├── templates/
│   └── index.html       # Model comparison, feature importance chart, prediction form
├── static/
│   └── style.css        # Dark theme styling
└── requirements.txt

My Other ML Projects

Project	Description	Repo
Diabetes Classifier	Model comparison + feature scaling demo	diabetes-classifier
Heart Attack Predictor	End-to-end classification pipeline	heart-attack-predictor
Medical Cost Predictor	Regression — what drives healthcare costs	medical-cost-predictor
Stock Price Predictor	LSTM — 5,884 tickers + crypto	stock-price-predictor

Author

@manny2341