F-RouND: Fog-based Rogue Nodes Detection in Vehicular Ad hoc Networks
FF-RouND: Fog-based Rogue Nodes Detection inVehicular Ad hoc Networks
Anirudh Paranjothi ∗ , Mohammed Atiquzzaman ∗ , Mohammad S. Khan † School of Computer Science, University of Oklahoma, Norman, Oklahoma, USADepartment of Computing, East Tennessee State University, Johnson City, Tennessee, USAe-mail: ∗ { anirudh.paranjothi, atiq } @ou.edu, † [email protected] Abstract —Vehicular ad hoc networks (VANETs) facilitatevehicles to broadcast beacon messages to ensure road safety.The rogue nodes in VANETs broadcast malicious informationleading to potential hazards, including the collision of vehicles.Previous researchers used either cryptography, trust values, orpast vehicle data to detect rogue nodes, but they suffer fromhigh processing delay, overhead, and false-positive rate (FPR).We propose fog-based rogue nodes detection (F-RouND), a fogcomputing scheme, which dynamically creates a fog utilizingthe on-board units (OBUs) of all vehicles in the region forrogue nodes detection. The novelty of F-RouND lies in providinglow processing delays and FPR at high vehicle densities. Theperformance of our F-RouND framework was carried out withsimulations using OMNET++ and SUMO simulators. Resultsshow that F-RouND ensures 45% lower processing delays, 12%lower overhead, and 36% lower FPR at high vehicle densitiescompared to existing rogue nodes detection schemes.
Index Terms —VANETs, rogue nodes, fog computing.
I. I
NTRODUCTION
Vehicular ad hoc networks (VANETs) are considered asthe keystone of intelligent transportation systems (ITS) toenhance road safety by reducing the number of accidentsand optimizing the traffic flow. VANETs provide vehicle-to-vehicle (V2V), and vehicle-to-infrastructure (V2I) commu-nication depends on dedicated short range communication(DSRC), which consists of a set of protocols for transmittingthe messages between vehicles and between vehicles and theroadside infrastructures, such as roadside units (RSUs), etc.[1, 2]. The vehicles are equipped with on-board units (OBUs)for transmitting and receiving the messages, including beaconmessages. Beacon messages are broadcasted in VANETs todisseminate network state or emergency information, whichcan be utilized to reduce road accidents and traffic congestion[3]. However, rogue vehicles, also known as rogue nodesbroadcast malicious information, such as false congestioninformation and collision warning by broadcasting low-speedvalues in beacon messages to change the normal behavior ofthe vehicles may lead to catastrophic consequences, such asthe collision of vehicles [4]. Detecting rogue nodes plays acrucial role in establishing a secure VANETs environment.Previous authors used either cryptography, trust values, orpast vehicle data to detect rogue nodes. Al-Otaibi et al. [5]presented a cryptography-based intrusion detection scheme(IDS) using fog computing. The proposed scheme considers RSUs as fog nodes for rogue nodes detection. However, theapproach [5] encounters a high processing delay and overheadin detecting rogue nodes when the RSUs are overloaded ornot available in the region. Zaidi et al. [6] proposed an IDS todetect rogue nodes based on past vehicle data. Each vehicleutilizes its OBU to detect false data propagated by the roguenodes. Ahmad et al. [7] proposed a trust-based scheme termedas trust evaluation and management (TEAM). The TEAMframework consists of three different trust models: entity-oriented, data-oriented, and hybrid-oriented trust models fordetecting rogue nodes. RSUs are used to compute trust scoresand segregates the vehicles based on the calculated trust score.The existing approaches [6, 7] have limitations, such as highdelay, overhead, and false-positive rate (FPR).To address the shortcomings of the existing rogue nodesdetection schemes, we introduce an OBU-based dynamic fogcomputing technique called fog-based rogue nodes detection(F-RouND). The F-RouND framework employs a two-foldprocess in rogue nodes detection: first, we use the conceptof guard node to detect rogue nodes. The guard node is thevehicle that has more neighboring vehicles in its transmissionrange, dynamically creates a fog utilizing the OBUs of allvehicles in the region, and then the dynamic fog is usedto compare the received beacon messages from all vehiclesto detect rogue nodes. Second, the guard node performsthe hypothesis test to validate whether the rogue nodes arecorrectly identified or not. Upon successful validation, theguard node broadcasts the information of rogue nodes to allvehicles in the region. We adopt fog computing, as it offersunique services, including low latency and high bandwidthcompared to traditional communication techniques [8].The difference between the F-RouND framework and ex-isting schemes is, each vehicle uses either its OBU or RSUto detect rogue nodes [5-7]. RSUs are deployed only in thecritical regions of the road. The absence of RSUs yields highprocessing delay and FPR. OBUs of an individual vehicleis highly resource-constrained encounters a high delay inanalyzing the data at high vehicle densities. Whereas, in theF-RouND framework, the guard node combines OBUs of allvehicles in the region in creating the dynamic fog. UtilizingOBUs of all vehicles increases the computational power ofdynamic fog resulting in low processing delay and FPR.Our objective is to reduce latency, increase true positive rate(TPR), and decrease FPR in detecting rogue nodes at high a r X i v : . [ c s . N I] F e b ehicle densities. We considered three existing rogue nodesdetection schemes for comparison: Fog-IDS [5], IDS [6], andTEAM [7]. The performance of our framework was carriedout using OMNET++ and SUMO simulators. Our resultslead to an exciting conclusion that the F-RouND frameworkreduces the latency and FPR, and performs up to 38% betterthan the existing rogue nodes detection schemes [5-7].The contributions of the paper are: 1) We proposed aframework that uses statistical techniques and traffic models todetect rogue nodes in VANETs with low delay, low FPR, andhigh TPR. 2) We introduced the guard node in the F-RouNDframework, which uses an OBU-based fog computing tech-nique to compare and validate the received beacon messagesfrom all vehicles in the region. 3) The proposed frameworkdoes not depend on any roadside infrastructures, includingRSUs in rogue nodes detection.The rest of the paper is structured as follows: related work isdiscussed in Section II. The proposed solution for rogue nodesdetection is presented in Section III. Section IV evaluates theperformance of our approach through extensive simulation.Finally, conclusions and future work are given in Section V.II. R ELATED W ORK
This section presents an overview of the most recentexisting schemes that detect rogue nodes in VANETs. Arshadet al. [9] proposed a beacon-based trust scheme to detect falsemessages in VANETs. Initially, the trust values of all vehiclesare assigned to be 0, and then based on the correctness of thedata, positive or negative trusts are assigned. The calculatedtrust of any vehicle reaches a predefined threshold limit areknown as rogue nodes, and then the information is broadcastedto all the vehicles in the region. However, the proposed work[9], suffers from high packet loss ratio (PLR) and FPR.Sedelmaci et al. and Ahmed et al. [10, 11] proposed trust-based schemes to detect rogue nodes. RSUs are used tocompute trust scores and detect rogue nodes based on thecalculated trust score. Yang et al. [12] proposed a tree-basedmachine learning algorithm to classify whether received datais valid or not based on the historical vehicle data. Theresults of the classification are then combined to detect roguenodes broadcasted false information. The frameworks [10-12]encounter a low TPR and high FPR in detecting rogue nodes.Zhang et al. and Shams et al. [13, 14] illustrated therogue nodes detection mechanism based on the support vec-tor machine (SVM) to resist false messages. The proposedmechanism [13] consists of a local trust module and a vehicletrust module, where the local trust module uses an SVM-based classifier to detect false messages, and the vehicle trustmodule uses SVM to derive comprehensive trust value forall vehicles. Finally, the results of both the local module andthe trust module are then combined to find rogue nodes inthe region. The authors [14] use the SVM-based module toanalyze the past vehicle data. Based on the analysis, the trustvalues of the vehicle are calculated, which in turn used fordetecting the rogue nodes. However, these approaches [13, 14]have high delay and overhead at high vehicle densities.
Fig. 1. Execution scenario of the F-RouND framework in the presence of arogue node using fog computing technique.
To overcome the limitations of the existing rogue nodesdetection schemes [5-7, 9-14], we propose the F-RouNDframework, which uses fog computing technique to detectrogue nodes in VANETs. The proposed framework does notdepend on either trust score or past vehicle data and performsbetter even when there are 40% rogue nodes in the region.III. P
ROPOSED
F-R OU ND F
RAMEWORK
In this section, we provide the working principle of ourproposed F-RouND framework. The F-RouND frameworkengages fog computing technique to detect rogue nodes due tothe fog can be created at the proximity of users and performscomputations at the edge of the network [8]. In order to detectrogue nodes, F-RouND employs the concept of guard node.The vehicle which has a greater number of the neighboringvehicle in its transmission range will act as a guard node.Guard node dynamically creates a fog using OBUs of allvehicles to compare and analyze vehicle speeds to detect roguenodes. All the vehicles in the region broadcast similar speedvalues as they are in similar traffic conditions and dependenton other vehicles under all circumstances. Thus, if there is asignificant difference in vehicle speed, the guard node consid-ers the vehicles as rogue nodes, and then the hypothesis test iscarried out to validate whether the rogue nodes are correctlyidentified or not. If the hypothesis test yields speed valueswithin the acceptance range, then the vehicles are consideredas honest nodes. Otherwise, the vehicles are highlighted asrogue nodes. The guard node broadcasts the information ofrogue nodes to all vehicles in the region to ignore the beaconmessages received further from the rogue nodes. One suchscenario of our F-RouND framework is depicted in Fig. 1.The computation power of the guard node increases whenthe number of vehicles increases as the OBUs of all vehiclesare utilized in creating a dynamic fog results in lower delaycompared to [5-7] rogue node detection schemes.
A. Selection of Guard Node
Rogue nodes are the vehicles broadcasting low-speed valuesto change the normal behavior of the vehicles for ownenefits. The guard node analyzes received beacon messagesfrom all vehicles to detect rogue nodes. The following threeassumptions are made in the selection of the guard node: first,we assume that the center vehicle has a greater number of theneighboring vehicles in its transmission range compared tothe front and tail-end vehicles. Hence, we select the centervehicle as a guard node. Second, we assume that the guardnode is the most trustworthy vehicle in the network. Thus,the guard node cannot be turned out to be a rogue node underany circumstances. Third, we assume that the total number ofvehicles ( N ) in the region at any given time is at least twoas the guard node needs at least two vehicle data to compareand analyze the beacon messages to detect rogue nodes.The guard node of the F-RouND framework is selected asfollows: initially, we take the mean of position vectors of allvehicles (i.e., P , P , ...., P N ) to find a unique center point ζ ,and then we calculate Euclidean distance between ζ and theposition vector of each vehicle to determine the point that hasthe minimum distance from ζ . Finally, the vehicle located atthis point will be selected as the guard node, G veh . ζ = 1 N N (cid:88) i =1 P i (1) G veh = arg min P i ∈ X (cid:107) ζ − P i (cid:107) (2)Where, X = { P , P ...., P N } . B. Speed and Density of Vehicles
The vehicles broadcast beacon messages every 100 ms. Inthe F-RouND framework, Greenshield’s mathematical modelis utilized to model the traffic flow in the region. Greenshieldtraffic model is considered to be a fairly accurate and simplemodel for real-world traffic flows works under the assumptionof density ( ρ ), and the speed of the vehicles ( S ) is negativelycorrelated [7]. The density can be calculated as: ρ = B msg · N (3)Where B msg is the beacon message broadcasted fromone vehicle id and N is the total number of vehicles inthe region. As the speed and density of the vehicles arenegatively correlated, the density increases when the speed ofthe vehicles decreases in the region. The relationship betweenspeed and density can be defined as: S = S max − ρρ max S max (4)Where S max is the speed of the vehicle when density is zeroand ρ max is the maximum density, also point at which speedof the vehicles becomes zero. In addition to usual parameters,such as speed, acceleration, braking status, location, gap, VIN,etc., the beacon message of all vehicles in the F-RouNDframework also includes the density information ( ρ ). TABLE IT
YPES OF E RROR AND D ECISIONS IN N ULL H YPOTHESIS T ESTING
Null hypothesis ( H )True FalseNull hypothesis ( H )decision Accept No error Type II error(False negative)
Reject
Type I error(False positive) No error
The selected guard node (Section 3A) creates a dynamic fogto compare and analyze the vehicle speed in beacon messagesto detect rogue nodes in the region. Once the rogue nodes areidentified, the guard node calculates the average density ( ρ avg )and the average speed ( S avg ) as follows: ρ avg = 1 N N (cid:88) i =1 ρ i (5) S avg = 1 N N (cid:88) i =1 S i (6)The guard node uses the average speed ( S avg ) to performthe hypothesis test. During the hypothesis test, vehicles thatcorrespond with average speed are termed as honest nodes. Incase the average speed difference is high or low, the upper andlower bound values are calculated to decide whether a receivedspeed value should be accepted or not. The hypothesis testprovides a significant contribution in reducing FPR at highvehicle densities compared to [5-7] schemes. C. Hypothesis Test to Validate the Speed of Vehicles
Hypothesis testing allows a confidence interval to the rangeof values that allows us to accept a claim with a certain con-fidence. The F-RouND framework performs a hypothesis testwith the speeds received from all vehicles in the region, whichallows the guard node to accept the speeds with a certainconfidence. Moreover, hypothesis testing is a commonly usedstatistical technique when we have two different claims, ofwhich only one claim can be true at any given time. In theF-RouND framework except for the guard node, we have twodifferent claims for all the vehicles in the region, i.e., eitherthe vehicle is honest or rogue. If the vehicle is honest, theguard node accepts the data, else if the vehicle is rogue, theguard node rejects the data, and then the information of therogue nodes is broadcasted to all the vehicles in the region.We use the hypothesis test to validate if the vehicle is honestor rogue using speed values in the beacon messages.There are two hypotheses involved in the hypothesis testingapproach: the null hypothesis ( H ) and the alternate hypoth-esis ( H a ). The null hypothesis is the claim that needs to betested, and the alternate hypothesis is everything else. If thenull hypothesis is accepted, then the alternate hypothesis isrejected, and vice versa. In the F-RouND framework, the nullhypothesis ( H ) is that the speed value received is from anhonest vehicle, and the alternate hypothesis ( H a ) is that thespeed value received is from a rogue node. Two types of ig. 2. Hypothesis test of F-RouND framework based on the average vehiclespeed to determine acceptance range values. error associated with the hypothesis testing approach: the firsttype of error (Type I error) occurs when the null hypothesisis wrongly rejected, also known as a false positive, andthe second type of error (Type II error) occurs when thenull hypothesis is wrongly not rejected, also known as afalse negative, as shown in Table I. False negative is not assevere as false positive as it may not lead to any catastrophicconsequences in the network.We use standard deviation ( σ ) to measure the variation ofaverage speed with speed values received from all vehicles. σ = (cid:118)(cid:117)(cid:117)(cid:116) N N (cid:88) i =1 ( S avg − S i ) (7)A low standard deviation indicates that the speed valuesreceived from the vehicles tend to be close to the averagespeed calculated by the guard node, while a high standarddeviation indicates that the speed values received from thevehicles have highly deviated from the average speed. Theupper and lower limits of our acceptance region will be S avg − σ and S avg + σ . The speed values received from the honestnodes fall in the acceptance region S avg − σ < S avg < S avg + σ when the null hypothesis is true, as shown in Fig. 2. Thus,the speed values received outside the acceptance region arerejected (i.e., S avg − σ > S avg > S avg + σ ).Unlike existing rogue nodes detection schemes [5-7, 9-14], the F-RouND framework works efficiently for all vehicledensities as well as for all road conditions. For example, incase of an accident or high dense downtown regions, the speedof all vehicles drop, which will decrease the average speed( S avg ) for the region, and as a result, the speed values ofthe honest nodes remain in the acceptance region. Once therogue nodes are identified, the guard node includes the roguenode id and the result of the hypothesis test, i.e., either 0 or1 (Eqn. 8) to the beacon messages, and then broadcast thebeacon messages to all vehicles in the region. All vehicles inthe region start ignoring the beacon messages received furtherfrom the rogue nodes to contain the damage. Rlt = (cid:40) S avg − σ < S avg < S avg + σ Otherwise (8)
TABLE IIP
ARAMETERS USED IN S IMULATION OF
F-R OU ND F
RAMEWORK
Parameters Values
Road length 3 MilesNumber of vehicles 500-4000Number of lanes 2Vehicle speed 30-65 Miles/hrBeacon message size 256 bytesTransmission range 500 mTechnique used Fog computingProtocol IEEE802.11pSimulator used Omnet++, SUMO
D. F-RouND Algorithm
Algorithm 1
F-RouND - Rogue nodes detection algorithm
Input: G veh receives B msg from all vehicles in the region Output: G veh broadcasts information of rogue nodes if ( N ≥ then Calculate ζ and Euclidean distance Assign G veh else GoTo 19 end if G veh dynamically creates a fog G veh receives B msg from all vehicles in the region for each B msg received do Calculate S avg and ρ avg Perform hypothesis test if S in the acceptance range then Declare the vehicle as honest node else
Declare the vehcile as rogue node
Store the rogue node id end if end for G veh broadcasts rogue nodes information through B msg Terminate the rogue nodes detection algorithmIV. P
ERFORMANCE E VALUATION
This section evaluates and analyzes the performance of ourF-RouND framework discussed in Section 3.
A. Analysis of F-RouND Framework
In this analysis, we calculated the probability of failure.Failure of the system can occur due to loss of connectivity ora resource, insufficient capacity of fog, etc. The probabilityof system failure P sysfail is calculated by: P sysfail = N,t max (cid:88) i =0 (cid:18) N, t max i (cid:19) d if (1 − d f ) N,t max − i (9)Where N is the number of vehicles, t max is the maximumtime taken by the vehicles to get connected, and d f is theprobability of success in the fog. Like quality of service(QoS), the probability of system failure contributes to the ig. 3. Comparison of the F-RouND framework with Fog-IDS, IDS, and TEAM schemes: (a) data processing time, (b) PLR, (c) average throughput, (d)overhead, (e) TPR, (f) FPR. performance of the F-RouND framework. A minimum numberof failures leads to the maximum performance of the fog. B. Simulation Setup
The main objective of our simulation is to evaluate theperformance of the F-RouND framework in the presence ofrogue nodes (Section 3). We used OMNET++ and SUMOsimulators to carry out the simulations. SUMO provides atrace of vehicular movements for a map imported from Open-StreetMap, while OMNET++ provides realistic modules, suchas the packet loss model, node deployment model, etc. forrealistic network behavior. We imported the city of Norman,Oklahoma, using OpenStreetMap into the SUMO simulator togenerate vehicle traces. The output of the SUMO simulator,i.e., the trace of vehicles, is given as input to the OMNET++simulator for rogue nodes detection. To assess the scalabilityand performance of the F-RouND framework, we performeda simulation with up to 4000 vehicles and 40% rogue nodes.Table II summarizes the parameters used in the simulation.
C. Performance Metrics
The simulations were performed based on the equationsformulated in Section 3. We considered the following metricsto evaluate the performance of the F-RouND framework and tocompare our results with Fog-IDS, IDS, and TEAM schemes: • Data processing time: The time needed by the guardnode to compare and analyze vehicle speed in the beaconmessages to detect rogue nodes in the region. • PLR: The ratio of the number of lost packets to the totalnumber of packets sent across a communication channel. • Average throughput: Average rate of successfully broad-casted beacon messages across a communication channel. • Overhead: The additional information exchanged be-tween the vehicles to detect rogue nodes in the region. • True positive rate: The percentage of rogue nodes isaccurately detected and classified as rogue nodes.TPR = No. of rogue nodes detected correctlyTotal no. of rogue nodes (10) • False positive rate: The percentage of honest nodes isincorrectly detected and classified as rogue nodes.FPR = No. of honest nodes detected incorrectlyTotal no. of honest nodes (11)
D. Simulation Results1) Data processing time:
The data processing time increasesas the number of vehicles increases as more time needed toprocess the vehicle speed of all vehicles, as shown in Fig. 3a.In the F-RouND framework, the computation power of theguard node increases when the number of vehicles increases inthe region as the OBUs of all vehicles are utilized in creatinga fog results in 45% lower processing time at high vehicledensities compared to [5-7] frameworks. In the 4000 vehiclessimulation, the data processing time is 43%, 52%, and 57%lower than Fog-IDS, IDS, and TEAM schemes, respectively.The evaluation of data processing time shows the F-RouNDframework is scalable and can handle high vehicle densities.
2) PLR:
The PLR is calculated against the number ofvehicles and increases for all schemes when the number ofvehicles increases from 500 to 4000. An increase in thenumber of vehicles increases the load on the network. Whenthe network hits maximum capacity, packet drops occur. Also,PLR increases due to the collision of some packets. In the-RouND framework, the high computation power of thefog resulting in an optimum network capacity even with anincreasing number of vehicles resulting in low PLR, i.e., 5%PLR at high vehicle densities, as shown in Fig. 3b.
3) Average throughput:
The average throughput of theF-RouND framework is calculated against the number ofvehicles, as shown in Fig. 3c. In the F-RouND framework,due to the high scalability of our dynamic fog (Section 3) andlow PLR (Fig. 3b), the number of successfully broadcastedmessages in the network increases with an increase in thenumber of vehicles resulting in high average throughput at allvehicle densities. In the 4000 vehicles simulation, the averagethroughput is 9%, 17%, and 23% lower than Fog-IDS, IDS,and TEAM schemes, respectively.
4) Overhead:
The overhead of the F-RouND frameworkincreases with the increasing number of rogue nodes as an ex-tensive hypothesis test needed to detect all rogue nodes in theregion (Section 3C). Overhead is the additional informationexchanged between the guard node and all other vehicles inthe region to detect rogue nodes. In the F-RouND framework,vehicle speed in the beacon message is used to detect roguenodes. Thus, the overhead of our F-RouND framework is 12%lower compared to [5-7] schemes, even when the number ofrogue nodes increased up to 40%, as shown in Fig. 3d.
5) TPR:
The F-RouND framework identifies rogue nodescorrectly (i.e., 100%) up to 35% rogue nodes in the region,as shown in Fig. 3e. As discussed in Section 3, the detectionof rogue nodes is two-fold: first, the guard node compares thereceived speed values of all vehicles to detect rogue nodesin the region. Second, once the rogue nodes are identified,the guard node performs a hypothesis test to validate if therogue nodes are correctly identified or not. However, when thenumber of rogue nodes is more than 35%, the TPR decreasesmarginally to 99%. It is difficult to detect the rogue nodewhen the speed varies gradually. However, to generate eithera false congestion scenario or catastrophic consequences, thetarget rogue node decreases the speed values suddenly. Thus,the F-RouND framework can detect rogue nodes even at highvehicle densities resulting in lower TPR compared to [5-7].
6) FPR:
The increase in FPR deteriorates the performanceof the proposed rogue node detection schemes. In the F-RouND framework, the rogue nodes detection relies onlyon the vehicle speed in beacon messages broadcasted by allvehicles in the region without using any trust scores or pastvehicle data. Moreover, validation of the rogue nodes usinghypothesis test results in 36% lower FPR compared to existingschemes [5-7] even when the number of rogue nodes increasesby up to 40% in the region. For a network with 40% roguenodes, the FPR is 31%, 51%, and 38% lower than Fog-IDS,IDS, and TEAM schemes, respectively, as shown in Fig. 3f.V. C
ONCLUSIONS AND F UTURE W ORK
We studied challenges in rogue nodes detection, such ashigh processing delay, high network overhead, poor resourceutilization, high FPR, and low TPR, notably when the numbernumber of rogue nodes increases at high vehicle densities. To address these problems, we proposed an OBU-based fogcomputing technique, called F-RouND which ensures 45%lower processing delay, 36% lower FPR, and 12% loweroverhead at high vehicle densities compared to existing roguenodes detection schemes [5-7]. We have analyzed the dataprocessing time, PLR, average throughput, network overhead,TPR, and FPR, and performed a simulation using OMNET++and SUMO simulators. Results showed that the F-RouNDframework is efficient, scalable, and performs up to 38% betterthan [5-7] schemes even when the number of rogue nodesincreases by up to 40% in the region. Moreover, the F-RouNDframework does not depend on any roadside infrastructureslike RSUs or trust scores or past vehicle data in rogue nodesdetection, which is a major advantage compared to existingrogue nodes detection schemes. In the future, we plan toextend this work on vehicular social networks platforms.This can be done by simulating the environment of thesocial networks and then detecting the malicious informationbroadcasted using rogue nodes detection techniques.R
EFERENCES[1] H. Hasrouny, A. E. Samhat, C. Bassil, and A. Laouiti, “VANET securitychallenges and solutions: a survey,”
Vehicular Communications , vol. 7,pp. 7-20, 2017.[2] A. Paranjothi, M. S. Khan, R. Patan, R. M. Parizi, and M. Atiquzza-man, “VANETomo: A congestion identification and control scheme inconnected vehicles using network tomography,”
Computer Communica-tions , vol. 151, pp. 275-289, 2020.[3] G. Loukas, E. Karapistoli, E. Panaousis, P. Sarigiannidis, A. Bezemskij,and T. Vuong, “A taxonomy and survey of cyber-physical intrusiondetection approaches for vehicles,”
Ad Hoc Networks , vol. 84, pp. 124-147, 2019.[4] R. G. Engoulou, M. Bella¨ıche, S. Pierre, and A. Quintero, “VANETsecurity surveys,”
Computer Communications , vol. 44, pp. 1-13, 2014.[5] B. Al-Otaibi, N. Al-Nabhan, and Y. Tian, “Privacy-preserving vehicularrogue node detection scheme for fog computing,”
Sensors , vol. 19, pp.965-983, 2019.[6] K. Zaidi, M. B. Milojevic, V. Rakocevic, A. Nallanathan, and M.Rajarajan, “Host-based intrusion detection for VANETs: A statisticalapproach to rogue node detection,”
IEEE Transactions on VehicularTechnology , vol. 65, pp. 6703-6714, 2015.[7] F. Ahmad, V. N. Franqueira, and A. Adnane, “TEAM: A trust evalu-ation and management framework in context-enabled vehicular ad-hocnetworks,”
IEEE Access , vol. 6, pp. 28643-28660, 2018.[8] A. Paranjothi, U. Tanik, Y. Wang, and M. S. Khan, “Hybrid-Vehfog:A robust approach for reliable dissemination of critical messages inconnected vehicles,”
Transactions on Emerging TelecommunicationsTechnologies , vol. 30, pp. 359-375, 2019.[9] M. Arshad, Z. Ullah, M. Khalid, N. Ahmad, W. Khalid, D. Shahwar,and Y. Cao, “Beacon trust management system and fake data detectionin vehicular ad-hoc networks,”
IET Intelligent Transport Systems , vol.13, pp. 780-788, 2018.[10] H. Sedjelmaci, S. M. Senouci, and M. A. Abu-Rgheff, “An efficientand lightweight intrusion detection mechanism for service-orientedvehicular networks,”
IEEE Internet of Things Journal , vol. 1, pp. 570-577, 2014.[11] S. Ahmed, S. Rubeaai, and K. Tepe, “Novel trust framework forvehicular networks,”
IEEE Transactions on Vehicular Technology , vol.66, pp. 9498-9511, 2017.[12] L. Yang, A. Moubayed, I. Hamieh, and A. Shami, “Tree-based intel-ligent intrusion detection system in internet of vehicles,”
IEEE GlobalCommunications Conference (GLOBECOM) , pp. 1-6, 2019.[13] C. Zhang, K. Chen, X. Zeng, and X. Xue, “Misbehavior detection basedon support vector machine and Dempster-Shafer theory of evidence inVANETs,”
IEEE Access , vol. 6, pp. 59860-59870, 2018.[14] E. A. Shams, A. Rizaner, and A. H. Ulusoy, “Trust aware support vectormachine intrusion detection and prevention system in vehicular ad hocnetworks,”