Machine Learning Based Trust Management framework for Sustainable Vehicular Networks

Abstract

Vehicular Edge Computing (VEC) augments resource-constrained devices in vehicular networks by bringing services and processing power closer to the end-user at the network edge. However, over time, these devices get overbur dened as a result of incoming requests, and network performance degrades. This can be addressed by utilizing nearby idle nodes and assigning tasks to nearby vehicles; a process known as computational task offloading. However, the presence of malicious nodes might put the entire network at risk; tasks cannot be offloaded if the node is untrustworthy; therefore service trustwor thiness is critical. The majority of the work on task offloading has focused on resource optimization, and the trustworthiness of services has received less attention. The traditional trust models focus on the aggregation of both direct and indirect observations. The choice of optimal weights for different factors in traditional approaches is still a key challenge given the dynamic nature of VANETs (vehicular ad hoc networks). Thus, we employ the estab lishment of trust and identification of malicious vehicles as a classification problem for task offloading in vehicular networks. We have simulated multi ple attacking patterns and generated a novel data set to identify misbehaving nodes. We trained multiple machine learning models on the generated data set. LSTM (Long Short Term Memory) reported the highest performance gain. Moreover, we deployed the trained model on edge nodes and pro posed a multi-criteria task offloading framework in vehicular networks. In the presence of adversary nodes in the network, the proposed task offloading framework with integrated intelligent layer outperformed baseline techniques in terms of task efficiency, effectiveness, and black hole failures.