PhishNet-FL: Federated Learning for Phishing Detection

PhishNet-FL is a federated learning system designed to detect phishing websites while preserving user data privacy. Using Flower for federated learning and TensorFlow for deep learning, PhishNet-FL enables multiple clients to collaboratively train a phishing detection model without sharing raw data.

This approach enhances security by distributing the learning process while still benefiting from shared knowledge.

🚀 Features

✔ Federated Learning (FL): Clients train locally and only share model updates, ensuring data privacy.
✔ Phishing Detection Model: A neural network trained on the UCI Phishing Websites Dataset.
✔ Decentralized Training: No need to centralize sensitive data, reducing security risks.
✔ Scalable Architecture: Supports multiple clients communicating with a central federated server.
✔ Easy Deployment: Simple client-server setup with minimal dependencies.

---

📂 Repository Structure

/PhishNet-FL
│── server.py          # Federated learning server
│── client.py          # Federated learning client
│── README.txt         # Project documentation
│── requirements.txt   # Dependencies
│── .gitignore         # Ignore unnecessary files

---

⚡ Installation

Before running the project, install the required dependencies:

pip install -r requirements.txt

---

🖥 Running the Server

The server manages the federated learning process by coordinating clients and aggregating their updates.

python server.py

You should see output confirming the server is running and waiting for clients to connect.

---

👨‍💻 Running Clients

Each client trains the phishing detection model on its local dataset and communicates updates to the server.

python client.py <client_id> <total_clients> <server_ip>

• <client_id> → Unique ID for the client (e.g., 0, 1, 2...)
• <total_clients> → Total number of clients participating in training
• <server_ip> → IP address of the federated learning server

For example, if you have 3 clients and your server is running on 192.168.1.10, start each client as follows:

python client.py 0 3 192.168.1.10
python client.py 1 3 192.168.1.10
python client.py 2 3 192.168.1.10

---

🧪 Model Architecture

The phishing detection model is a deep neural network built using TensorFlow:

• Input Layer: 30 features (preprocessed phishing website attributes)
• Hidden Layers:
  • Dense (64 neurons, ReLU activation)
  • Dropout (20%)
  • Dense (32 neurons, ReLU activation)
  • Dropout (20%)
  • Dense (16 neurons, ReLU activation)
• Output Layer: 1 neuron (Sigmoid activation for binary classification)

---

📊 How Federated Learning Works in PhishNet-FL

1️⃣ Server Initialization: The federated server starts and waits for clients to connect.
2️⃣ Client Training: Each client trains a local model on its dataset without sharing raw data.
3️⃣ Model Updates: Clients send model updates (weights) to the server instead of raw data.
4️⃣ Federated Aggregation: The server averages the received model updates and improves the global model.
5️⃣ Global Model Distribution: The updated global model is sent back to clients for the next training round.
6️⃣ Repeat: The process repeats for multiple rounds until the model converges.

This ensures privacy-preserving, decentralized machine learning across multiple devices.

---

📌 Dependencies

• flwr (Flower for federated learning)
• tensorflow (Deep learning framework)
• scikit-learn (Data preprocessing and evaluation)
• ucimlrepo (Dataset retrieval)
• numpy (Numerical computations)

---

📚 Future Improvements

🔹 Support for additional datasets and real-time phishing detection.
🔹 Advanced privacy-preserving techniques like differential privacy.
🔹 More efficient aggregation methods for improved performance.

---

📬 Contact & Contribution

💡 Found a bug? Have a feature request? Feel free to open an issue or contribute!
📧 Reach out if you have any questions or need help setting up PhishNet-FL.

Happy training! 🚀🔐