ShodhPatra: International Journal of Science and Humanities

FMDP: Federated Learning-Driven Model for DDoS Attack Detection and Prevention in Vehicular Edge Computing

Vehicular Edge Computing (VEC) enables low latency processing of data by deploying computational resources at the edge of the network for intelligent transportation systems. VEC does have a significant vulnerability to Distributed Denial of Service (DDoS) attacks, which often target the Road Side Units (RSUs), as by attacking RSUs this can disrupt vehicular communication and various components of the system’s reliability. We propose a scalable DDoS detection framework that preserves privacy using Federated Learning and Long Short-Term Memory (LSTM) neural networks. The proposed architecture is designed in a layered fashion containing three layers, Vehicle, Edge (RSU) and Cloud. RSUs are designed to host lightweight sample LSTM models that will classify network traffic in real time. In order to judge the degree of the attack we have also defined an attacks degree measurement that aims to quantify irregular traffic flows based on network analysis statistics and entropy, which may also allow for tweeting filters early on in the DDoS attack life cycle. In order to preserve data privacy and scalability, we adopted Federated Learning that allows RSUs to train their own models, while sharing only model updates to the central model aggregator, the central model maintains overall distributive learning system consistency between all RSUs. We used Python to simulate the system on one-thousands samples with four-hundred samples of malicious attack. The detection accuracy of the system was found to be 92.0% detections accuracy with detection rate was 93.2% with 6.8% failures. Compared to other traditional model methodologies like DoSRT, our approach demonstrated superior real time performance, scalability and adaptibility in a VEC environment.