Protocol Misbehavior Analysis using Multivariate Statistical Analysis and Machine Learning in Vehicular Ad Hoc Networks
Kumar Sharshembiev, Seong-Moo Yoo and Elbasher Elmahdi
We propose a novel framework to analyze 802.11p physical and MAC layer parameters’ threshold values during the broadcast in Vehicular Ad Hoc Networks (VANETs) in order to detect misbehaving nodes and prevent broadcast storms. This low overhead threshold analysis model is based on the 802.11p / WAVE incoming and outgoing data statistics to detect if the node is behaving properly. To avoid misjudgment and false alarms, the accuracy of the estimate of the threshold value is validated using the autoencoder neural network. Our method and detection technique are based on the IEEE 802.11p MAC protocol and weighted p-persistence routing protocol. The main contribution of this paper is proposing analytical and experimental models to accurately detect initial stages of protocol misbehavior using the optimal threshold analysis to trigger the mitigation mechanism for misbehaving nodes in VANETs. For each proposed model, we get results obtained from simulations and the validity of the models is confirmed by observing the close relationship between the results.
Keywords: vehicular ad hoc networks, autoencoder neural network, misbehaving node, principal component analysis, Mahalanobis distance