Detection of DDOS attacks in Vehicular Ad Hoc Networks using Ensemble Deep Learning Model and Optimization Technique

• The proposed architecture secures the VANETs by utilizing an SDN controller and an ensemble deep learning model. • It involves the segregation of control planes and data planes in alignment with SDN architecture. The AlexInception-BiLSTM-AttentionNet model addresses the limitations of standard Self-Organizing Maps while enhancing their efficacy by integrating two levels of clustering and classification. • A new SDN controller architecture is proposed for the evolved nodes at base stations. This architecture is designed to optimize network performance and enhance the network's security against DDoS attacks. • To further optimize the performance of the proposed scheme, a metaheuristic algorithm inspired by the foraging behavior of dung beetles called Dung Beetle Optimization (DBO) is used. • The proposed model achieves 99.31 % accuracy, and 99.28 % specificity for UNSW NB15 dataset, 98.94 % accuracy, and 99.13 % specificity for Litnet20 dataset. Vehicular Ad-Hoc Networks (VANET) stands vital in integrating intelligent transport services but equally remains vulnerable to DDoS attacks because of the inherent dynamic and wireless routing. This paper proposes a secure VANET architecture with SDN and an ensemble deep learning model-AlexInception-BiLSTM-AttentionNet-to increase detection level as well as ensure an immediate security response. This architecture supports layered operation that decouples the control plane from the data plane of VANETs through the help of SDN controllers. Further, it brings in a novel SDN controller framework design for edge base stations. To this end, a two-layer classification and clustering scheme is piled on top of metaheuristic nature of Dung Beetle Optimization (DBO) to speed up convergence and fine-tune model parameters. Two benchmark datasets, UNSW_NB15 and Litnet20, were used in comparison with the ton of literature-based state-of-the-art models. This method recorded promising results: 99.31 % Accuracy, 99.28 % Specificity on UNSW_NB15 and 98.94 % Accuracy and 99.13 Specificity on Litnet20. The results illustrate the ability of the proposed model to robustly identify DDoS attacks with a high precision. Having a serious ability of adaptation, scaling, and executing, therefore, our SDN-driven VANET security model is suggested to be an insightful and efficient platform to monitor the traffic in real-time and work for its mitigation against some attacks via the next-generation intelligent transportation system.