Analysis of BayesNet Classifier for DDoS Detection in Vehicular Networks

In vehicular networks, Distributed Denial of Service (DDoS), an extension of Denial of Service (DoS) attacks, disrupt the normal vehicular network services. Existing DDoS detection systems in vehicular networks use a variety of Machine Learning (ML) methods. These techniques claim to be relevant and successful in attack detection in vehicular networks. The ML-based solutions use DDoS dynamics like adaptive traffic, heterogeneity, and packet features etc. As a result, a question arises whether these ML-based techniques are capable of handling novel security threats. In this paper, we use BayesNet technique on different datasets to test its performance and analyse the results to validate its usage for protecting against novel network security threats. BayesNet helps in decision making by improving data management and analysis. We use two datasets: simulation-based dataset and CIC-DDoS-2019 dataset. The CIC-DDoS-2019 dataset has 12 subsets and we use SYN, UDPLag and Portmap data subset. These datasets contain three types of DDoS attacks, i.e., SYN-flood attack, UDPLag attack, and Portmap attack. We test BayesNet to see the performance of the algorithm in detecting the mentioned forms of DDoS. We consider the parameters of accuracy, True Positive Rate (TPR), and False Positive Rate (FPR) for performance evaluation. In all the results, we observe that accuracy is always higher than 90%. TPR is very high, i.e. weighted TPR is 0.99 and wighted FPR of all tests is 0.076 that is quiet low. All these results prove that BayesNet ML technique is capable of detecting DDoS attack forms significantly.