- Yuvaraj K
- Nithish S
- Krishnaa S
DDoS attacks have become the most popular means of targeting and attacking websites in recent times due to the large-scale attention that they draw towards the general public. While not considered a concrete cybersecurity threat, as they don’t cause any real permanent damage to websites or result in any leakage of data, they are a complete nuisance that prevents actual users from accessing the services offered by websites and applications. As such, it is important to be able to predict or detect such attacks in advance in order to better prepare for and tackle them in an efficient manner.
Here, we introduce FEDGuard, a Federated Learning system that is decentralized and distributed, enabled to aid in the detection of DDoS attacks. This system keeps on learning and improving from new network flow data that is localized and independent, while aggregating the results of the local training and testing with a global master. It is efficient and privacy-focused, ensuring that localized data remains secure, while delivering a powerful detection system with a high and consistent accuracy rate of over 90%. It implements an efficient preprocessing mechanism and feature selection using the ANOVA F-value metric for the network flow data, and is flexible in the number of clients that it can support for decentralized learning, wherein it uses standard machine learning classification algorithms for binary classification, such as Logistic Regression.
FEDGuard represents a novel approach in tackling the issues of DDoS attack detection, ensuring reliability and accuracy in an optimized manner.