Loan Approval Prediction System

A binary classification neural network built with Keras to predict loan approval decisions based on applicant financial and demographic data. Uses deep learning techniques to assist in credit risk assessment.

Features

• Neural Network Classification: Multi-layer perceptron for binary prediction
• Keras API: Clean and intuitive model building with Keras Sequential API
• Feature Engineering: Comprehensive data preprocessing and normalization
• Model Persistence: Trained model saved as HDF5 for deployment
• Performance Metrics: Accuracy, precision, recall, and confusion matrix analysis
• Interactive Notebook: Step-by-step Jupyter notebook with visualizations

Tech Stack

• Framework: TensorFlow/Keras
• Language: Python 3.8+
• Libraries:
  • Pandas & NumPy for data manipulation
  • Scikit-learn for preprocessing and metrics
  • Matplotlib/Seaborn for visualization
• Model: Saved as Timothy_Balch_project_loan_model.h5

Installation

Prerequisites

• Python 3.8+
• Jupyter Notebook
• pip package manager

Setup

1. Clone the repository:

git clone https://github.com/CodeBalch25/Loan-predictions-using-keras.git
cd Loan-predictions-using-keras

2. Install dependencies:

pip install tensorflow keras pandas numpy scikit-learn matplotlib seaborn jupyter

3. Launch Jupyter:

jupyter notebook "Loan predictions using Keras API.ipynb"

Project Structure

Loan-predictions-using-keras/
├── Loan predictions using Keras API.ipynb    # Main notebook
├── Timothy_Balch_project_loan_model.h5      # Trained model
└── README.md                                 # Documentation

Model Architecture

model = Sequential([
    Dense(64, activation='relu', input_shape=(n_features,)),
    Dropout(0.3),
    Dense(32, activation='relu'),
    Dropout(0.2),
    Dense(16, activation='relu'),
    Dense(1, activation='sigmoid')
])

model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy', 'precision', 'recall']
)

Features Used

• Applicant Income: Annual income of the primary applicant
• Coapplicant Income: Income of co-applicant (if applicable)
• Loan Amount: Requested loan amount
• Loan Term: Duration of the loan (months)
• Credit History: Binary indicator of credit history (1 = good, 0 = poor)
• Property Area: Urban, Semi-Urban, or Rural
• Education: Graduate or Not Graduate
• Self Employed: Yes or No
• Marital Status: Married or Single
• Dependents: Number of dependents

Data Preprocessing

1. Handle Missing Values: Imputation with median/mode
2. Encode Categorical Variables: One-hot encoding for categorical features
3. Feature Scaling: StandardScaler for numerical normalization
4. Train-Test Split: 80/20 split with stratification
5. Class Balancing: Handle imbalanced dataset if needed

Model Performance

• Accuracy: ~85%
• Precision: High precision to minimize false positives
• Recall: Balanced recall for fair lending practices
• F1 Score: Harmonic mean of precision and recall

Usage

Training

# Load and preprocess data
X, y = load_and_preprocess_data()

# Build model
model = build_loan_model(input_dim=X.shape[1])

# Train
history = model.fit(
    X_train, y_train,
    validation_split=0.2,
    epochs=100,
    batch_size=32,
    callbacks=[early_stopping, model_checkpoint]
)

# Save model
model.save('loan_model.h5')

Prediction

# Load trained model
model = load_model('Timothy_Balch_project_loan_model.h5')

# Predict on new data
predictions = model.predict(X_new)
loan_approved = predictions > 0.5

Evaluation Metrics

• Confusion Matrix: Analyze true/false positives and negatives
• ROC Curve: Receiver operating characteristic
• Precision-Recall Curve: Trade-off analysis
• Feature Importance: Identify key predictors

Key Insights

1. Credit History: Most significant predictor of loan approval
2. Income Ratio: Combined income to loan amount ratio matters
3. Property Area: Urban areas show higher approval rates
4. Education: Graduate status positively correlated with approval

Neural Network Benefits

• Non-linear Relationships: Captures complex patterns in data
• Automatic Feature Learning: Reduces manual feature engineering
• Scalability: Handles large datasets efficiently
• Continuous Improvement: Model can be retrained with new data

Regularization Techniques

• Dropout: Prevents overfitting (30% and 20% dropout rates)
• Early Stopping: Stops training when validation loss stops improving
• Batch Normalization: Stabilizes learning process

Future Enhancements

• Add SHAP values for model explainability
• Implement hyperparameter tuning (GridSearch/RandomSearch)
• Add more features (employment history, debt-to-income ratio)
• Build REST API for real-time predictions
• Create web interface for loan application processing
• Add fairness metrics to ensure unbiased lending
• Implement ensemble methods (stacking, bagging)

Business Impact

• Faster Decisions: Automated approval process
• Reduced Bias: Data-driven decisions
• Risk Assessment: Better credit risk evaluation
• Cost Savings: Reduced manual review time
• Scalability: Handle thousands of applications

Compliance & Ethics

• Ensure Fair Lending Act compliance
• Audit for demographic bias
• Maintain transparency in decision-making
• Regular model monitoring and updates

Technologies

• TensorFlow/Keras: Deep learning framework
• Pandas: Data manipulation
• Scikit-learn: Preprocessing and metrics
• NumPy: Numerical operations
• Matplotlib/Seaborn: Visualization

Contributing

Contributions welcome! Submit a Pull Request.

License

MIT License

Author

Timothy Balch - @CodeBalch25

Acknowledgments

• TensorFlow team for Keras API
• Scikit-learn community
• Financial ML community for best practices

Tags

deep-learning keras classification financial-ml neural-network tensorflow credit-scoring machine-learning binary-classification loan-prediction