CS8.501: Research in Information Security

Trust-aware Authentication and Authorization for IoT Devices

Lightweight Flower-based Federated Learning project for trust classification using physical layer features. This repo implements a server and 3 clients, per-client preprocessing, dataset generation, model checkpoints, and evaluation utilities.

Requirements

• Python 3.10+ (used in development)

• Recommended packages: torch, flwr, pandas, scikit-learn, joblib, seaborn, matplotlib, numpy

• Use a virtualenv:

  python3 -m venv .venv
  source .venv/bin/activate
  pip install -r requirements.txt

Project structure

• federated_learning/

  • server.py — starts Flower server and evaluates global model each round
  • custom_strategy.py — FedAvgWithModelSaving
  • base_client.py — base client implementation used by client scripts
  • model.py — BaseModel, get/set weights utilities
  • dataset.py — TrustDataset wrapper
  • model_evaluation.py — evaluate saved checkpoints, produce confusion matrix & metric plots

• data_preprocessing/

  • generate_dataset.py — parse raw captures into per-distance CSVs
  • generate_trust_label.py — compute Trust Factor and map to Class labels
  • split_train_test_data.py — split, scale and write per-client train / global test CSVs
  • data/preprocessed/ — processed per-distance CSVs

Execution Steps

1. Activate your virtualenv

   • source .venv/bin/activate

2. Generate the dataset

   python3 -m data_preprocessing.generate_dataset
   python3 -m data_preprocessing.generate_trust_label
   python3 -m data_preprocessing.split_train_test_data

3. Start Flower server (run in its own terminal)

   • python3 -m federated_learning.server

4. Start client processes (one per client, separate terminals)

   • python3 -m federated_learning.client_script_name

5. After training, evaluate a saved checkpoint:

   • python3 -m federated_learning.model_evaluation artifacts/model_checkpoints/parameters_round_x.pkl --output-dir artifacts

Useful constants

Edit constants.py to change training behavior:

• NUM_ROUNDS, CLIENTS_REQUIRED, BATCH_SIZE, WINDOW_SIZE, SCALER_PATH, NUM_CLASSES