Federated Learning Intrusion Detection System
Developed and evaluated a privacy-preserving intrusion detection system using federated learning. Built a PyTorch multilayer perceptron and used Flower to compare centralised and federated training approaches on approximately 946,000 CICIDS2017 network flows.
- Centralised F1 score: 99.38%
- Federated F1 score: 99.20%
- Approximately 946,000 network flows
- Compared detection performance and privacy trade-offs
- Dissertation manuscript in preparation for academic publication
- Python
- PyTorch
- Flower
- Scikit-learn
- Pandas
- Docker
- CICIDS2017