🧠 Loan Repayment Prediction — Keras API Project

A machine learning project that builds a neural network using the Keras API to predict whether a borrower will repay a loan. The model is trained on historical lending data and evaluated using classification metrics.

---

📌 Overview

This project uses a deep learning classification model built with Keras (TensorFlow backend) to predict loan repayment outcomes. The workflow includes data preprocessing, feature scaling, model building, training, and evaluation.

The goal is to demonstrate practical usage of neural networks for real-world financial risk prediction.

---

🎯 Objective

Predict whether a borrower will:

• ✅ Repay the loan
• ❌ Default on the loan

Using borrower financial and credit-related features.

---

🧩 Tech Stack

• Python
• TensorFlow / Keras
• Pandas
• NumPy
• Scikit-learn
• Matplotlib / Seaborn
• Jupyter Notebook

---

📊 Dataset

The dataset contains borrower and loan-related attributes that may include:

• Income
• Credit history
• Loan amount
• Interest rate
• Employment length
• Debt ratio
• Payment history indicators

Target variable:

• loan_repaid (0 or 1)

---

🔬 Project Workflow

1️⃣ Data Preprocessing

• Load dataset
• Handle missing values
• Remove leakage features
• Encode categorical variables
• Feature scaling using StandardScaler / MinMaxScaler

---

2️⃣ Exploratory Data Analysis

• Distribution plots
• Correlation analysis
• Class balance check
• Feature relationship visualization

---

3️⃣ Model Building (Keras Sequential API)

Neural network includes:

• Dense layers
• Activation functions (ReLU / Sigmoid)
• Dropout for regularization
• Binary classification output layer

Example structure:

• Input Layer
• Dense Hidden Layers
• Dropout Layers
• Output Layer (Sigmoid)

---

4️⃣ Training

• Train/Test split
• Model compiled with:
  • Binary crossentropy loss
  • Adam optimizer
• Model trained over multiple epochs
• Validation monitoring

---

5️⃣ Evaluation

Model evaluated using:

• Accuracy
• Confusion Matrix
• Classification Report
• Precision / Recall / F1-score

---

📈 Results

classification_report:

               precision    recall  f1-score   support

           0       0.99      0.44      0.61     15658
           1       0.88      1.00      0.93     63386

    accuracy                           0.89     79044
   macro avg       0.93      0.72      0.77     79044
weighted avg       0.90      0.89      0.87     79044

---

▶️ How to Run

Clone repo

git clone https://github.com/rohitb281/keras-api-project.git
cd keras-api-project

---

Install dependencies

pip install tensorflow pandas numpy scikit-learn matplotlib seaborn

Open the notebook:

jupyter notebook

Run all cells in:

Keras-Project-loan-repayment.ipynb

---

🧠 Concepts Demonstrated

• Neural network classification
• Keras Sequential API
• Feature preprocessing pipeline
• Overfitting control with Dropout
• Model evaluation metrics
• Binary classification workflow

---

🚀 Possible Improvements

• Hyperparameter tuning
• ross-validation
• ROC curve & AUC analysis
• Feature importance analysis
• Model comparison with non-NN algorithms
• Deployment as API

---

⚠️ Notes

• Results depend on preprocessing choices
• Class imbalance handling can improve performance
• Feature leakage must be avoided for real deployment

---

📄 License

• Open for educational and portfolio use.

---

👤 Author

• Rohit Bollapragada
• GitHub: https://github.com/rohitb281