Secured Federated Learning for DDoS Detection in Heterogenous Telecom Cloud Networks using Recurrent Neural Networks

Abstract

The recent evolution of cloud computing has enabled the cloudification of Telecommunication (Telecom) network functions. The cloud-based Telecom infrastructure is more scalable, flexible, and cost-efficient for service providers. However, a significant security challenge for Telecom cloud providers is ensuring the availability of services provided to users by mitigating Distributed Denial of Service (DDoS) attacks. The fact that Virtual Network Functions (VNF) in the Telecom cloud are hosted on the Internet makes them easy targets for large-scale DDoS attacks. This study proposes the use of secured supervised Federated Learning (FL) with an efficient Hybrid Recurrent Neural Network (H-RNN) for DDOS attack mitigation in the Telecom cloud. The proposed H-RNN model combines LSTM, a Bidirectional GRU (BiGRU), and a Bidirectional LSTM (BiLSTM) to obtain a state-of-the-art LSTM+BiGRU+BiLSTM model. FL is used with Secure Sockets Layer (SSL) encryption, which supports data privacy and integrity in heterogeneous Telecom cloud networks. The simulation results using the CICDDOS2019 benchmark dataset displayed a detection accuracy of 99.59%, a False Positive Rate (FPR) of 0.042%, and an average detection time of 0.062 ms. A novel H-RNN model and secured FL are proposed to enable deep-learning-based anti-DDoS technology building and deployment in cloud-based Telecom networks.