Dynamic $\boldsymbol{K}$-Means Clustering for Sybil Attack Detection in VANETs

Detecting Sybil attacks in Vehicular Ad hoc Networks (VANETs) presents the challenge of validating proposed detection methods across varying scenarios because many proposed detection methods are not validated across diverse scenarios. Clustering also poses difficulties, as determining the optimal number of clusters$(k)$is a complex problem. In traditional clustering algorithms, the number of clusters ($\boldsymbol{k}$) can vary depending on the type of VANET scenario considered in determining the appropriate value of$\boldsymbol{k}$. To solve this problem of optimal selection of$k$, a dynamic$k$-means clustering framework, which adaptively determines the number of clusters required to group nodes and detect Sybil nodes with high accuracy, is developed. This method effectively clusters nodes in VANETs while identifying Sybil attacks. OMNET++ and SUMO simulators are used for simulation experiments. The proposed approach achieves an average accuracy of 99%, an average adjusted rand index of 0.91, and an average completeness of 0.76 across different scenarios and varying percentages of attackers on the network.