FCL-PRIDS: Federated Continual Learning for Privacy-Preserving Intrusion Detection

📖 Abstract

This repository implements FCL-PRIDS, a framework designed to secure Industrial IoT (IIoT) networks against evolving cyber threats. Unlike traditional centralized IDSs, this project employs Federated Continual Learning (FCL) to train a global model across distributed edge devices without sharing raw traffic data.

The core classification engine is a 1D-Convolutional Neural Network (1D-CNN), optimized for extracting spatial features from time-series network traffic (MQTT, CoAP, HTTP).

📂 Repository Structure

.
├── src/
│   ├── FCL_PRIDS_CNN.py            # The 1D-CNN model architecture
│   ├── train_federated.py          # FL training loop (Server/Client logic)
│   ├── train_centralized.py        # Baseline centralized training for comparison
│   ├── compute_communication_cost.py # Scripts to measure bandwidth usage
│   └── split_clients.py            # Data partitioning logic for non-IID settings
├── results/
│   ├── accuracy_comparison.png     # FCL vs Centralized performance
│   └── centralized_model.joblib    # Saved model weights
├── data/                           # (Excluded via .gitignore)
└── requirements.txt                # Python dependencies