2024-25
Privacy preserving DDoS detection using Federated Learning across Distributed Systems
Federated Learning based DDoS detection with data poisoning resistance.
About the Project
This project focuses on Federated Learning to maintain data privacy while collaboratively training DDoS detection models across distributed systems. Further we implement data poisoning resistance with byzantine client tolerance.
Key Features
- →Uses Flower Architecture for Federated Learning
- →Tolereant to Byzantine clients with data poisoning attacks
- →30 features considered from a total of 88 features [CICDDoS2019 Dataset]
- →Dockerized for easy deployment and scalability
- →An easy to understand attack detection and monitoring dashboard
FLDDoS monitoring dashboard
What I Learned
Preprocessed the CICDDoS2019 dataset,sampled dataset from 23.9gb to 2.24gb[50k records],reduced features from 88 to 30+2, balanced dataset 25k as benign and 25k as attack, implemented the federated learning setup using Flower framework and compared multi-krum with hybrid data-poisoning resilent aggregation techniques.