Blockchain-Empowered Clustered Federated Learning IDSs

🚀 Project Snapshot

A lightweight demo that blends clustered federated learning with a hash-anchored model ledger to detect network intrusions.

1. Each client trains locally on normal traffic from the UNSW-NB15 dataset.
2. Updates are merged at cluster heads via a coordinate-wise median (IFCA).
3. Cluster heads are averaged into a global model.
4. Every weight vector’s SHA-256 hash is appended to a file-based ledger (optionally anchored on-chain with Solidity).

---

🐍 Requirements

Category	Packages / Tools	Notes
Python	3.12 or newer	Core runtime
Core ML stack	PyTorch ≥ 2, NumPy, pandas, scikit-learn	Required for training & evaluation
Optional plotting	matplotlib	Only needed if --plot-file is used
On-chain anchor	web3, py-solc-x, local Ethereum node (Ganache / Hardhat / Anvil)	Needed only when pushing model hashes on-chain

📈 IFCA in a Nutshell

1. Broadcast – push every cluster head to the sampled clients.
2. Local training – each client fine-tunes all heads for a few SGD epochs on its normal-only shard.
3. Head selection – the client keeps the head with the lowest loss and uploads only that update.
4. Robust aggregation – the coordinator applies a coordinate-wise median to merge updates within each cluster.
5. Global averaging – refreshed heads are averaged into a single global model and its SHA-256 hash is logged.
6. Optional fine-tune – run a short, class-weighted SGD pass on a balanced mixed server buffer to regain anomaly visibility.

🧾 Model Ledger

Mode	What Happens	How to Use
File-based (default)	SHA-256 hash of every cluster/global weight vector is appended to ledger/model_ledger.jsonl.	No extra setup; enabled out-of-the-box.
On-chain anchor (optional)	Hashes are stored in an Ethereum smart contract (ModelLog.sol) for immutable auditability.	1. Start a local node (Ganache / Hardhat / Anvil). 2. Export creds:   bash<br>export PROVIDER_URL=http://127.0.0.1:8545<br>export DEPLOYER_ADDRESS=0x...<br>export DEPLOYER_KEY=0x...<br> 3. Deploy contract:   bash<br>python ledger/deploy.py<br> 4. Switch ledger/logger.py to send transactions instead of writing to disk.

🤝 Contributing

We happily welcome pull requests, feature ideas, and bug reports!

• Fork the repo and open a PR for enhancements or fixes.
• Issues are the place to ask questions or propose larger changes.
• Areas where help is most valuable:
  • Smarter or more robust aggregation strategies
  • Balanced dataset generation / preprocessing scripts
  • Hardening the model-ledger for production (IPFS, L2, etc.)

Feel Free to reachout to me Gmail : anilkumarbarupal.01@gmail.com

Project is released under the IIT License—have fun hacking! 🚀