A Survey of Attacks and Detection Mechanisms on Intelligent Transportation Systems: VANETs and IoV

Abstract

Vehicular ad hoc networks (VANETs) have become one of the most promising and fastest growing subsets of mobile ad hoc networks (MANETs). They are comprised of smart vehicles and roadside units (RSU) which communicate through unreliable wireless media. By their very nature, they are very susceptible to attacks which may result in life-endangering situations. Due to the potential for serious consequences, it is vital to develop security mechanisms in order to detect such attacks against VANETs. This paper aims to survey such possible attacks and the corresponding detection mechanisms that are proposed in the literature. The attacks are classified and explained along with their effects, and the solutions are presented together with their advantages and disadvantages. An evaluation and summary table which provides a holistic view of the solutions surveyed is also presented.

Full text

Abstract—Vehicular ad hoc networks (VANETs) have become one of the most promising and
fastest growing subsets of mobile ad hoc networks (MANETs). They are comprised of smart vehicles
and roadside units (RSU) which communicate through unreliable wireless media. By their very
nature, they are very susceptible to attacks which may result in life-endangering situations. Due to
the potential for serious consequences, it is vital to develop security mechanisms in order to detect
such attacks against VANETs. This paper aims to survey such possible attacks and the
corresponding detection mechanisms that are proposed in the literature. The attacks are classified
and explained along with their effects, and the solutions are presented together with their advantages
and disadvantages. An evaluation and summary table which provides a holistic view of the solutions
surveyed is also presented.
Keywords— Attacks, intrusion detection, misbehavior detection, IoV, VANET, security
1. INTRODUCTION
Vehicular Ad Hoc Networks (VANETs) are a special type of mobile ad hoc network used for
communication among and between vehicles and roadside units. VANETs are an emerging technology for
many applications, including congestion monitoring and traffic management. For example, vehicles on a
road where an accident has occurred can alert each other to take an alternative route in order to avoid the
traffic jam that has built up following the accident. Beside safety-related applications, there are also other
applications such as infotainment, payment services, insurance calculations based on usage, and other
similar means. These are applications which require vehicles to communicate with infrastructure, people
and the Internet, resulting in VANETs having evolved into the universal paradigm known as the Internet of
Vehicles (IoV) [1].
The special characteristics of VANETs, such as high mobility, dynamic network topology, and
Fatih Sakiz* and Sevil Sen
Department of Computer Engineering, Hacettepe University, Turkey
A Survey of Attacks and Detection Mechanisms
on Intelligent Transportation Systems:
VANETs and IoV
* Corresponding author. Tel.: +90-312-297-7500
E-mail addresses: fatihsakiz@hacettepe.edu.tr, ssen@cs.hacettepe.edu.tr.

predictable node movements, require new algorithms and protocols to be developed specific to this new
environment. Security also poses a challenge, since it may affect life-or-death decisions. To date, studies
have focused on VANET technologies with limited attention on security.
One of the first surveys on security attacks against VANETs found in the literature is proposed by Isaac
et al. in [2]. The authors summarize the general approaches against attacks and report that for VANETs,
many security challenges still remain unresolved. Since the survey’s publication in 2010, studies on
detection and prevention mechanisms have accelerated, with many approaches having been proposed. In
this research, a detailed analysis of attacks against VANETs is presented, together with the detection
systems proposed to date for each attack type. Furthermore, response mechanisms that are proposed for
preventing or minimizing damage to the system are covered. Some solutions originally developed for
MANETs are also included in the survey, since it is believed they could also be applicable to VANETs.
1.1. Security Challenges in VANETs
Mobile ad hoc networks introduce new security issues which should be taken into account: lack of central
points, mobility, wireless links, cooperativeness, and lack of a clear line of defense [3]. The specific
characteristics of VANETs make these issues more challenging besides, introducing the following new
issues:
Privacy: It is difficult to provide user security while at the same time respecting privacy. Taking into
consideration that authorities may need information from vehicle drivers in case of an event, and that
drivers may want to keep their information (identity, location history, etc.) protected helps to understand
the trade-off between user privacy and security [4]. In order to prevent the tracking of all vehicles (‘big
brother’ scenario), a system should both provide users with anonymity, while also enabling the possible
determination of a user’s real identity upon legitimate requests from the appropriate authorities (police,
manufacturers, courts, etc.) [5].
Scalability: The number of vehicles worldwide is estimated at over 1 billion; a number that continually
increases [6]. In addition, the number of vehicles connected to VANETs is expected to exceed 250 million
by 2020 [7]. Currently, there is no global authority providing security for such networks, largely because it
is a challenging task to define standardized rules for VANETs since the aforementioned privacy-security
trade-off differs from one country to the next [8]. In order to facilitate this, worldwide coordination
between local authorities would be needed in order to provide standardized security.
Mobility: The topology of VANETs changes very rapidly due to one-time interactions between vehicles.
While nodes can be observed moving at a maximum speed of 20 m/s in MANET simulations [9], the
speeds of vehicles are in fact much faster than this limit. As a result, link breakages between vehicles are a
common occurrence in vehicular networks. In particular, links between vehicles driving in opposite
directions only last for a few seconds; hence the network can frequently become disconnected. Vehicles

generally communicate with each other for just a short time, and then never see each other again, which
hardens reputation-based systems. However, the topology changes in a more predictable way as opposed to
MANETs simulations found in the literature.
Hard-delay constraints: Many applications, particularly safety-related applications in VANETs, require
real-time responses. If these requirements are not met, the consequences could be catastrophic through
incidents such as accidents or delayed rescue operations. Furthermore, these real-time requirements make
applications vulnerable to Denial of Service (DoS) attacks. Some researchers state that many safety-related
applications should focus primarily on the prevention of attacks, rather than detection and recovery because
of real-time demands [10]. However real-time attack detection is also critical in such applications,
especially when insiders bypass existing prevention mechanisms.
Cooperativeness: Many of VANETs algorithms and protocols assume that data will be disseminated by
vehicles in communication. This feature makes vehicular networks vulnerable to attacks such as bogus
information attacks. Many security mechanisms also rely on the cooperativeness of vehicles, since local
data might not be sufficient for the prevention and detection of attacks.
Mobile ad hoc networks consist of various devices which have different computational and storage
capacities from hand-held devices to powerful laptops. In addition, such devices usually run on battery
power. Therefore limited resources are an issue faced by MANETs which should be considered while
designing security solutions. However, the nodes in vehicular ad hoc networks are either vehicles or stable
roadside units (RSUs) which have sufficient energy and computing power. Therefore VANET nodes will
not be an easy target for energy depletion attacks such as Sleep Deprivation Torture [11] in MANETs.
Moreover, we could introduce more security features for VANETs which might be impracticable for
resource-constrained MANETs.
Another security challenge for MANETs which has less impact on VANETs is lack of a clear line of
defense. Even though vehicular ad hoc networks do not have central points where security mechanism can
be placed as in wired networks, roadside units could play a key role in security by carrying out resource-
intensive jobs such as collecting alarms raised by vehicles and making decisions. Furthermore, these units
and vehicles could be more protected and secure than hand-held devices in MANETs. RSU-based security
solutions such as RSU-aided certificate revocation and RSU-based intrusion detection have already been
proposed in the literature. On the other hand, largescale deployment of RSUs is a costly approach.
There have been many security solutions proposed for MANETs over the past decade. While some of
these approaches are also adaptable to VANETs, most are deemed unsuitable for these highly dynamic
systems due to the aforementioned reasons. Therefore, new approaches or adaptations of existing
approaches to VANETs are needed. In this study, the focus is placed on attacks against vehicular
communication systems, and the solutions proposed in the literature in order to detect such attacks and
attackers. Even though there are existing prevention mechanisms, they may be susceptible to unknown

attacks. Furthermore, those mechanisms cannot protect the network from insider attackers with authorized
system access. In this study, the main focus is on detection mechanisms.
1.2. Evolution from VANETs to IoV
The Internet of Things (IoT) has been an emerging paradigm. IoT consists of different types of devices
and technologies in order to provide a connection among things at any time, from any place, to any
network. IoT has attractive areas of use such as smart home systems, assisted living, smart energy, e-health,
and intelligent transportation systems. There has been a large increase in the number of things connected to
the Internet. According to Gartner [12], Internet of Things will grow to 26 billion devices/units by 2020. It
is expected that a considerable number of these devices will be vehicles, which form the Internet of
Vehicles (IoV) or the Internet of Cars. IoV evolved from VANETs and is expected to eventually evolve to
become the Internet of Autonomous Vehicles [13].
There are many open research areas on the Internet of Things, from identification and communication
technologies to standardization [14]. Since many of the devices which constitute IoT are not designed with
security in mind, security attacks and solutions are also some of the main research challenges. In this paper,
attacks against IoV are classified in two groups, based on the target location of the attackers:
1. Inter-vehicle attacks: Vehicles could obtain valuable information from other vehicles or from the
environment to provide functionalities such as traffic congestion detection or systems for deceleration
warning. For instance, vehicles could exchange useful information such as accident notifications, traffic
congestion, and road conditions in order to assist in traffic management. Therefore, misbehaving nodes and
falsified data sent by these nodes in such critical applications could lead to such drastic results as a loss of
life or loss of energy and money. VANETs are vulnerable to new forms of attack, from dropping attacks to
bogus information attacks. Furthermore, since vehicles are connected through wireless communication
links, they are also susceptible to eavesdropping and traffic analysis attacks. Even though VANETs share
the vulnerabilities of wired networks and ad hoc networks such as spoofing and denial of service, they have
additional security challenges due to their very nature, including dynamic but predictable topology changes
and delay-tolerant data dissemination.
2. Intra-vehicle attacks: Intra-vehicle communication describes communications within a vehicle.
Modern road vehicles have a swarm of sensors for checking the road condition, vehicle distance, obstacle
detection, fire detection, vehicle speed/acceleration sensors, message display system, and an On-Board Unit
(OBU) which consists of vehicle-to-vehicle and road-to-vehicle communication systems, among others.
Intra-vehicle attacks such as deceiving a sensor/system could damage the vehicle and the environment. For
example, an attack which spoofs GPS information or disables the steering or braking system in an
autonomous vehicle could be extremely dangerous [13]. Furthermore, with the proliferation of the Internet
of Things, vehicles will be more susceptible to attacks and malwares infected from the Internet, and a sub-

system of a vehicle could be compromised and become remotely controllable. However, these types of
attacks are out of scope for this study.
In this study, the primary focus is inter-vehicle attacks. Studies on such attacks and solutions have
accelerated in recent years. In addition, inter-vehicle communications are considered more challenging than
intra-vehicle communications, since vehicular communications are provided both when vehicles are
stationary and mobile [15]. It is predicted that studies on intra-vehicle attacks will accelerate with the use of
IoT and Internet of Autonomous Vehicles in the near future.
The remainder of the paper is organized as follows: attacks against VANETs are classified in Section 2,
and proposed solutions for each attack type are given in Section 3. Section 4 presents the general evaluation
of the proposed solutions, and conclusions are drawn in Section 5.
2. ATTACK TYPES
With the proliferation of VANETs, new security risks against these highly mobile, but predictable
networks are highly likely to be exploited. In this section, existing attacks against VANETs are classified
according to their goals and methods. Attacks that are very specific to some routing protocols are not
studied in the literature, hence are not considered in this study. While some of the attacks are derivations of
existing attacks against MANETs, some of them are specific to VANETs.
2.1. Sybil Attack
Sybil attack can be classified as one of the most dangerous attacks in VANETs. In a Sybil attack
scenario, a node (vehicle) can pretend as if it has more than one identity. In other words, other nodes in the
network are unable to distinguish if the information originates from one vehicle or from more than one
vehicle. The main aim of the attacker is to shape the networks based on his/her goals. For example, an
attacker could manipulate other vehicles’ behaviors such as making them take a different road from their
scheduled route. Besides being one of the most dangerous forms of attack, Sybil attack is also among the
most difficult to detect [16]. It becomes more risky on networks using geographical routing, since the
attacker claims that the vehicle is in several positions by sending incorrect information about its position.
Furthermore, it could show events occurring in positions other than their genuine positions.
One type of Sybil attack is called a Node Impersonation Attack. In VANETs, each vehicle in the network
has a unique identity and vehicles use their identities while communicating with other vehicles in the
network [16]. However, if a vehicle changes its identity without the knowledge of the RSU or the network,
it could introduce itself as a different vehicle as in a Sybil attack. For example, a vehicle involved in a
traffic accident could change its identity to appear as a moving vehicle in the network. Hence, other
vehicles in the network see this vehicle as a different vehicle from those involved in the accident. Then, the
malicious vehicle could send incorrect information about the road conditions to the surrounding RSUs.

2.2. Denial-of-Service Attack (DoS)
Denial-of-Service (DoS) attacks aim to make valid activities of a system unavailable. The attackers
mainly send far more requests than the system can handle. In VANETs, an attacker could try to shut down
the network established by RSUs, and stop communication between vehicles and/or RSUs [16]. As a result
of a DoS attack, attackers cannot communicate with each other, and vehicles do not receive network
information such as road status, resulting in severe consequences. In a Distributed Denial-of-Service
(DDoS) attack, nodes could launch an attack from different locations, thereby making any detection harder.
Nodes launching a DDoS attack could aim to harm not only the vehicles in the network, but also RSUs,
which are an important aspect of the infrastructure in VANETs.
There are various types of DoS attacks in VANETs. JellyFish, intelligent cheater, and flooding attacks
are some known examples to be found in the literature. Aad et al. [17] present JellyFish attack, which is a
general class of protocol-compliant DoS attack against MANETs. It follows all routing protocol
specifications, unlike many other types of routing attacks. An attacker could disorder, delay, or periodically
drop packets it was supposed to forward. Eventually it exploits vulnerabilities of end-to-end congestion
control protocols in order to drastically decrease network performance. JellyFish attack could easily be
inherited by VANETs.
Similar to the JellyFish attack, intelligent cheater attack [18] also remains unsuspicious by following
routing protocol specifications. The attacker appears to be operating normally for most of the time, but in
fact just misbehaves in a discontinuous manner. Intelligent cheater attack and JellyFish attacks could easily
bypass trust mechanisms. Because of their sneaky nature it is very difficult to detect such attacks, requiring
end-to-end control mechanisms with long term monitoring for their detection [17]. However, long term
monitoring could be impracticable for VANETs due to their highly mobile nature.
Flooding attacks generate traffic in order to exhaust network resources such as bandwidth, CPU, power,
and other similar means. Flooding attacks can be divided into two groups: data flooding and routing control
packets flooding. The consequences of each attack type are the same. Resources in the network become
unavailable to legitimate users. In a data flooding attack, an attacker could create useless data packets and
send them to all nodes through their neighbors. However, the attacker needs to first set the routes with all
possible nodes in the network. In a route request flooding attack, the attacker broadcasts route request
control packets to nodes which do not exist in the network.
Another type of DoS attack is the jamming attack, which refers to occupying the channel used in the
network by transmitting radio frequency signals consisting of illegitimate traffic. The attack could be
performed by an attacker who is not necessarily a member of the network. Since this paper’s focus is on
insider attacks and jamming is a general problem for wireless networks, solutions against jamming are out
of scope for this research.

Considering the fact that anyone with limited knowledge could perform DoS attacks, and could therefore
prevent vehicles from getting real traffic events, the impact and the likelihood of the attack is considered
very high. In addition, DoS attacks must be detected as quickly as possible and response mechanisms
activated on time, since it is very difficult for the network to respond once an attack has been successfully
performed. Besides detection and response mechanisms, mitigation techniques such as proposed by Biswas
et al. [19] could be employed.
2.3. Blackhole Attack
Whereas DoS and DDoS aim to shut down the network, another attack that shapes the network is the
Blackhole attack. It is known as a serious threat for MANETs and refers to an attacker that manipulates
other nodes into sending their packets through itself as much as possible. In VANETs, an attacker vehicle
could exploit routing protocols such as claiming that it has the best path for the destination vehicle/RSU or
that it is in the best position to forward the packets. By broadcasting false routing information, it makes
other vehicles prefer to send their packets via itself, assuming that it is on the true path. After misrouted
victim vehicles send their packets to the attacker, it generally just discards all packets intentionally and as a
result, packet losses occur in the network (Fig. 1).
: Attacker
: Accident
Notification
Fig. 1. Blackhole attack
Consequences of the attack for VANETs are more serious since packet losses in safety-related
applications could cause life-endangering accidents. In [20], an analysis on the performance of Blackhole
attacks in VANETs is given, and it is shown that the attack could affect the network in terms of end-to-end
delay, throughput and network load, and that AODV is more vulnerable to the attack than OLSR. Besides
simply dropping packets, attackers could also send packets between each other in such a way as to create
their own network. For example, when a route request comes to a malicious vehicle, it could send the route
request to another malicious vehicle. In that manner, information important to the network will not be
forwarded by the attacker and may not be sent to the other vehicles as malicious vehicles only

communicate between themselves, rather than the rest of the network. Therefore, vehicles other than the
two malicious vehicles would not receive broadcasted safety messages.
2.4. Wormhole Attack
A wormhole attack [21], as already known from MANETs, is generally performed by two or more
compromised nodes which involve themselves in as many routes as possible by advertising they know the
shortest path to any destination. The goal of the attacker is to modify logical topology of the network in
order to collect and/or manipulate large amounts of network traffic. In order to perform the attack in
VANETs, after receiving a packet which should be forwarded, an attacker vehicle encapsulates the packet
and sends it to another compromised vehicle (Fig. 2). The latter opens the encapsulated packet and spreads
it. Since the original packet is encapsulated during the transmission, the hop count field cannot be
increased, no matter how many hops are between them. Therefore, similar to the Blackhole attack, those
two malicious vehicles make routing protocols prefer the link between them as the best route to any
destination, instead of closer routes that already exist in the network. In addition, the attack could be
performed even without compromising a node. The attacker could simply record the traffic at one point in
the network and tunnel them through another via an out-of-band channel in order to replay or use it
somewhere else. As a result, important information sent through the tunnel may not be broadcast/unicast
[18], which can impose a considerable impact on communication. In another form of this attack, the
vehicles could create their own private network.
Wormhole
: Attacker
Fig. 2. Wormhole attack
2.5. Bogus Information Attack
In VANETs, vehicles use the information which is generated or forwarded by other vehicles or RSUs.
However, received information may not always be true. A vehicle could generate false information on its
own and then send it to the network [16]. The attacker generally aims to manipulate other vehicles with
selfish and/or malicious intent. For example, a vehicle may generate information about a fake accident on

the road and then send that information to other vehicles in order to make them take another road. This is
more effective when there is no other vehicle to verify that information and the attack is very difficult to
detect. Moreover, the consequences are more serious if there is an attacker moving around quickly – also
called motorway attacker [22] – and broadcasts bogus information to groups it encounters. Since each
group forms a separate network and does not know about the attacker’s criminal records within other
groups, the attacker could affect many vehicles without being detected. Even though a bogus information
attack could cause very high impact by causing changes to drivers’ behaviors, the attacker usually must
have adequate knowledge of the network to avoid statistical detection mechanisms which limits the
likelihood of the attack.
2.5.1 False Position Information
Disseminating false position information is a critical problem in VANETs, since safety-related
applications are heavily dependent on reliable position information. In addition, analysis of the effects of
false position information in VANETs shows that it could decrease the overall packet delivery ratio by up
to approximately 90% [23]. As a result, disseminating false position information could cause performance,
reliability, and security problems in VANETs.
2.5.2 Sensor Tampering
Since the OBU of a vehicle will probably be installed in a position with limited access, an attacker could
try to deceive sensors by simulating false conditions in order to provide expected outputs. That technique is
effective because such a deception will most likely remain unnoticed by an intra-vehicle detection system.
For example, by braking within short periods, an attacker could manipulate safety-related applications in
such a way as to make it appear that there is traffic on the road. Therefore, traffic jam messages will be
broadcast over the network. Sensor tampering also covers illusion attack and GPS spoofing.
2.5.2.1 Illusion Attack
Illusion attack is peculiar to VANETs. The attacker mainly exploits the human psychological intuition.
To do so, the attacker affects the behaviors of other drivers by disseminating false information in concert
with a scene [24]. Firstly, the attacker needs to realize or create a suitable traffic situation in order to
prepare the scene. Therefore, when other drivers receive corresponding false information messages, they
are more likely to believe in them. For instance, if there are a lot of cars moving slowly at the front of the
traffic they are in, drivers will probably believe that there is an accident ahead and consider alternative
routes after receiving false warning messages (which indicate an accident). Secondly, the attacker needs to
generate corresponding false message by deceiving its own sensor(s) in order to make them report valid but
false message(s) instead of modifying their output(s) by itself [2]. The messages will remain intact and
valid. As a result, false information could be distributed over the network. The attack is very difficult to
detect, even within the vehicle itself.

2.5.2.2 GPS Spoofing
This attack is also known as a tunnel attack [25]. An attacker could inject false position information to
another vehicle(s) by using GPS simulators. The victim could be waiting for a GPS signal after leaving a
physical tunnel or a jammed-up area. The GPS simulator could generate signals which are stronger than
original GPS signals. Therefore, even a vehicle which receives an original signal from the satellite will
prefer to accept false position information sent by the attacker.
2.6. Replay Attack
In VANETs, messages could be stored for reuse later in order to deceive other entities in the network, as
in MANETs. This is referred to as replay attack, and the aim is to exploit the conditions at the time when
the original message is sent. After gathering information that moves around the network, the attacker could
store that information and resend it to the network later on, even though it is no longer true or valid. In
addition, the attack could be performed by the original sender. For example, an attacker could save a
received message about an accident or traffic event which happened sometime in the past and then resend it
later on (Fig. 3). Until the message becomes expired, the attacker could easily reuse it to deceive others.
However, utilizing mechanisms which ensure the integrity of messages timestamps restricts its likelihood.
Time: T1 Time: T2
: Attacker
: Accident
Notification
: Attacker
: Accident
Notification
Fig. 3. Replay attack
2.7. Passive Eavesdropping Attack
Passive eavesdropping attack refers to monitoring the network to track vehicle movement or to listen in
on their communication by utilizing wireless medium characteristics. Malicious vehicles could simply
intercept and examine the messages which flow in the network. This passive attack is also known as traffic
analysis attack or stealth attack [26]. The goal of the attacker is to gather information about the vehicles and
communication patterns for further attacks. It is usually carried out before implementing other types of
attacks such as blackhole and DoS attacks. The impact of passive eavesdropping attack could be very high,
since it could be a part of an extremely sophisticated low-and-slow form of attack.

There are also other types of attacks such as route disruption attacks [27] in which attackers take
advantage of the vulnerabilities and the cooperativeness of routing protocols. However, in the literature it
was found that researchers mainly focus on routing protocol-independent attacks such as dropping, Sybil,
and such like. Furthermore, some attacks reported in the literature are explored for an application scenario,
especially for safety-related applications. While a few studies proposed especially for some routing
protocols, which are explicitly specified in this survey, some of them aim at detecting application-specific
attacks. At the top level, attacks can be classified according to network protocol stacks. All of the
aforementioned attacks and the corresponding layers in which they could perform are presented in Fig. 4.
Other classifications are also possible such as passive and active attacks, atomic and composite attacks, etc.
Fig. 4. Attacks with corresponding Internet Protocol Stack Layers
3. SECURITY SOLUTIONS
This section presents the security solutions proposed in the literature for each of the attack types
mentioned in Section 2.
3.1. Sybil Attack
Sybil attack was introduced by Douceur in 2002 [28] and is one of the most dangerous and evasive
attacks against VANETs since attackers could easily bypass security mechanisms based on honest majority
assumption by using multiple identities. Sybil attacks employed with DoS attacks are claimed to be among
the most dangerous forms of attack in VANETs [5]. Correspondingly, many researchers in the literature
focused on detecting Sybil attacks.
In one of the earliest solutions for Sybil attacks, Golle et al. [29] devised a heuristic approach called
adversarial parsimony in order to detect Sybil attacks in vehicular networks. Informally, it means finding
the best explanation for corrupted data received. After improving the nodes’ sensor capabilities such as
using cameras and exchanging data via a light spectrum in order to verify that a claimed position is true,
determining which nodes really exist is possible. Data collected from the sensors is used to reach
information that enables distinguishing between nodes. After exchanging that information between

vehicles, the heuristic mechanism detects inconsistencies in the case of a Sybil attack by comparing
received data to a VANET model maintained by each vehicle and refers all of its knowledge about the
VANET. However, detection mechanisms are neither supported with simulation nor explained in depth.
In another early study, Xiao et al. [30] presented a lightweight security scheme based on estimating the
position of a node by analyzing its signal strength distribution. In this approach, three roles are assigned to
vehicles. Claimer is the vehicle which periodically broadcasts beacon messages, including its identity and
position. Witness is the vehicle residing in the claimer’s signal range. Witness nodes measure signal
strengths of the claimer and then save this information together with the corresponding claimer’s identity
information in their memory. Thus they keep a neighbor list which will be broadcast within the next beacon
message. When a beacon message is received, verifier vehicles wait for a period of time in order to collect
previous measurements of the claimer from witness nodes. Then, they can calculate an estimated position
of the claimer. RSUs are used to certificate positions of vehicles around them in order to be sure about the
direction of a vehicle. Hence, information received from Sybil nodes could be ignored. Also, all witnesses
for a claimer should come from the opposite direction. Since vehicles from the same direction are potential
Sybil nodes, it is ensured that all witnesses are physical vehicles. However, an attacker can change the
strength of the signal to a value corresponding false position information and bypass the consistency check
mechanism. A study on the effectiveness of using signal strength distribution to detect the attack is given in
[31], and [32] shows that the type of antenna in use has a remarkable effect on detection.
Zhou et al. [33] propose a privacy-preserving detection scheme without forcing vehicles to disclose their
identities in the network. The Department of Motor Vehicles (DMV) assigns a unique pool of pseudonyms
instead of assigning exactly one ID for each vehicle and pseudonyms in the same pool are hashed to a
common value. Thus, a vehicle can use any pseudonym in its pool and preserve its privacy. Since the hash
is stored at the roadside units (RSU) and also at the DMV, a RSU could check whether received
pseudonyms belong to the same pool or not. When a RSU becomes suspicious of a Sybil attack, it sends the
pseudonyms and the hash values of the possible attacker to the DMV. After that, the DMV confirms
whether or not the suspicious pseudonyms have been assigned to the same vehicle. The last step is
necessary because the RSU could itself have been compromised by the attacker. In that case the RSU
would need to be revoked. Although the solution can detect Sybil attacks while achieving privacy, it
requires individual vehicles to be registered and administered by trusted authorities [34]. The authors have
improved their scheme by making it adaptive based on traffic volumes in [35].
Similar to [33] in terms of preserving privacy, Yong Hao et al. [36] devise a protocol that enables
vehicles to detect a Sybil attack in a cooperative manner. The protocol utilizes group signature to preserve
privacy and correlation of mobility traces. It includes three phases: probing, confirmation and quarantine.
In the probing phase, vehicles periodically broadcast their geographic information along with identifiers of
the vehicles around them. The confirmation phase refers that in the case of an anomaly detection, vehicles

around the possible attacker inform others by broadcasting warning messages with their partial signatures.
If the number of vehicles which report anomalies reaches a threshold, it is possible to derive a complete
signature and the attacker is identified. However, if no vehicle is able to derive a complete signature after
the possible attacker is inspected by the first n consecutive vehicles which come from the opposite
direction, the possible attacker is considered as benign. In the quarantine phase, the latest geographic
information of the Sybil node will be notified by the vehicles around. The identified attacker could be
isolated for a while or reported to legal authorities. According to simulations, the security protocol is
efficient; however it is not evaluated in the existence of multi attackers.
Lee et al. [37] introduce a DTSA (Detection Technique against a Sybil Attack) protocol which uses
Session Key based Certificate (SKC) and protects privacy of the vehicles. Firstly, each vehicle’s unique ID
should be registered to a global VANET server. Then, a vehicle V generates its anonymous ID and
validates it by using a local VANET server. Next, they both generate a session key and use it in order to
generate V’s local certificate. After that, V could transmit messages using its local certificate. Any
receiving vehicle could validate V’s true identity by asking a local VANET server to send V’s local
certificate and compare it to the received message. If the result is not a match, the attack is detected. Even
though simulation results show that detection time is low, using SKCs requires a large amount of overhead
data.
Similar to DTSA [37] in terms of infrastructure, Rahbari et al. [38] present a cryptographic method which
utilizes PKI infrastructure in order to detect a Sybil attack. The approach requires RSU support and
includes four phases. In phase 1, each vehicle has to be registered and receive its group authentication key
from RSU in order to start transmitting, receiving and authenticating messages within the group. In phase 2,
when RSU receives a message from vehicle V, it forwards the message to the Local CA since the private
key of a Local CA is necessary to decrypt the message. In phase 3, after decrypting the message, the Local
CA needs the private key of V in order to check the identity of V. Therefore the Local CA sends a request
to the Home CA. In phase 4, the Home CA replies with the private key of vehicle V and after that, the
Local CA is able to detect a Sybil attack by comparing the reply message received from the RSU.
Simulation results show that the proposed method has low delay and detect the attack efficiently. However,
as the authors stated, vehicles moving to regions which belong to other CAs cause problems in the
detection process.
Recently, Feng et al. [39] suggested a system called Event Based Reputation System (EBRS) in order to
detect Sybil attacks. Each vehicle has a public key and pseudonyms which are valid for a limited time and
validated by the Trusted Authority (TA) over RSUs, and a local certificate issued by an RSU. In addition,
each RSU stores the pseudonyms, locally generated session key and local certificate of each vehicle in its
area. When a vehicle V receives a warning message about an event E from a neighbor N, V will send N’s
pseudonym and N’s local certificate that is encrypted with N’s session key to local RSU. After that, the

RSU is able to find N’s session key corresponding to the received pseudonym and decrypt N’s local
certificate by using the key. Then, if the decrypted certificate matches with N’s local (previously stored)
certificate, N’s certificate is validated. Therefore, it is possible to detect forged, stolen, or expired
pseudonyms. Furthermore, V will store or update the reputation and trusted values of event E in its event
table. The system does not alert the driver of V unless both the reputation and trusted values of E reaches
predetermined corresponding thresholds. If the warning message about E received from a Sybil attacker is
false, the values would not increase since other vehicles have not reported the false event. According to
simulations, EBRS could effectively detect Sybil attacks. However, the assumption that RSUs and OBUs
could not be compromised may not always hold.
Another privacy-preserving solution called Footprint, which is based on similar motion trajectories of
vehicles, is presented by Chang et al. in [40]. A dedicated RSU infrastructure is deployed and managed by
a Trust Authority (TA) which is also responsible for trust relationship between entities in the network. An
anonymous vehicle V communicates with encountered RSUs and demands authorized messages as a proof
of presence located near to a specific RSU within a specific time. As V moves, consecutive authorized
messages collected from encountered RSUs form a trajectory of the vehicle. When V starts communicating
with a RSU and/or vehicle, it provides its trajectory (chain of authorized messages) to the other party to be
verified. Since the authorized messages can be used to track the vehicle, they must be anonymously signed
by RSUs. Therefore, instead of identifying the real trajectories of the vehicles, similarity between
anonymous trajectories of a possible Sybil group is used in the detection process. In addition, two
authorized messages issued by one RSU must be linkable only for a short period of time in order to prevent
an attacker revealing the trajectory of V. Consequently, vehicle V could detect a Sybil “community” by
comparing similarities between each pair of trajectories of vehicles. According to evaluation results,
Footprint has high detection rate. However, it requires a dedicated infrastructure formed by RSUs and it
makes an assumption that all RSUs are trustworthy.
Chen et al. [41] rely on the similarity of Sybil nodes’ motion trajectories, which are unrealistic and
unacceptable in the real world, assuming Sybil nodes have the same location and motion trajectories all the
time. The solution makes Sybil attack detection possible for each vehicle independently. It needs
authorized infrastructures in order to provide tamper-free periodic digital signatures to nearby vehicular
nodes. The authorized infrastructures could be RSUs and mobile units such as police vehicles and public
transport vehicles. Vehicular nodes carry these signatures (infrastructure, time) and the signatures are
exchanged between neighboring vehicles. With the limited assistance from the infrastructures, each node
could detect the Sybil attack independently in four steps. After gathering signature vectors of all the
neighboring nodes, it calculates the difference value of the gathered signature vectors. In judging step, a
threshold is used to determine whether or not a Sybil attack exists. Finally, by classifying, Sybil attackers
are detected and all Sybil nodes fabricated by the same malicious node are included into one Sybil set.

However, as shown in Grover et al. [42], if two or more vehicles driving in opposite directions receive the
same signature at almost the same time from a RSU and this situation continues for a time exceeding a
predefined threshold, those vehicles could be falsely identified as Sybil nodes (see Fig. 5 – Vehicle A and
Vehicle B).
Park et al. [34] propose another approach based on timestamp series without using a vehicular public key
infrastructure. Before one vehicle sends a message to another, it first sends a request to a nearby RSU in
order to obtain a timestamp for the message. After that, the message is sent. The timestamps could be used
to find each vehicle’s recent trajectory and time. If vehicles receive similar timestamp series from the same
RSUs for a certain amount of time (see attacker and Sybil vehicles in Fig. 5), they will be considered as
Sybil nodes and treated as a single node. However, similarly to Chen et al. [41], two vehicles coming from
opposite directions could be falsely accused as Sybil nodes because they will receive similar timestamps for
a short while. Grover et al. [42] presented a similar system based on the theory that two nodes cannot have
the same set of neighbors for a time longer than a threshold. The difference from the previous works in [34]
and [41] is to be able to detect Sybil attacks by only using records collected from neighbor nodes. It does
not ask for support from surrounding infrastructures such as RSUs. However, the false accusation problem
still remains unresolved.
Time: T1
RSU
A
RSU
Time: T2
A
: Attacker
: Sybil Vehicle
: Timestamp
: Digital Signature
: Attacker
: Sybil Vehicle
: Timestamp
: Digital Signature
Fig. 5. Motion trajectories over time
To sum up, most of the promising detection techniques described focus on observing similarity of motion
trajectories, which is the most significant characteristic of Sybil nodes in order to detect an attack.
However, considering that many people follow the same route to their place of work at almost the same
time, such as in the daily rush hour, false positive rates caused by false accusation of benign vehicles could
increase. Moreover, vehicles in a network could follow the same routes in different modes such as
platooning or as a motorcade during a ceremony; therefore, different models should be constructed for such
cases.

3.2. Denial-of-Service Attack
The effects of a DoS attack in VANETs might be more severe than expected. For example, it could bring
down the network and hence, cause traffic accidents. Some solutions, such as [17] that proposed detection
of DoS attacks against MANETs, could also be employed in VANETs.
Soryal et al. [43] present a solution to detect DoS attacks in IEEE 802.11 DCF. They use a Markov chain
model in order to generate an adaptive threshold, which is the maximum rate of messages any node can
send over time with respect to the number of other nodes. If a node notices the number of CTS (Clear-to-
Send) messages received per second for a destination address is above the threshold, the sender of the CTS
messages is tagged as an attacker. The simulation results show that the threshold-based approach is able to
detect the attacker(s) efficiently. However, the solution is not scalable because it is designed for nodes
communicating with a single hotspot which acts the same as a RSU.
Verma et al. [44] devise a system to prevent DoS attacks by monitoring TCP packets. The mechanism,
called the request detector, is deployed at the edge router of innocent hosts and can be considered as a
RSU. The request detector keeps a table which records source and destination IP addresses by using Bloom
filter, a space-efficient probabilistic data structure that guarantees no false negatives. The system
continuously monitors SYN and corresponding SYN-ACK packets and maps them. If the number of
outstanding SYN packets becomes more than the determined threshold within a certain amount of time or
an unmapped ACK/SYN message is received, a suspicious alarm is sent to The Response Detector which is
deployed at the protected master node. Therefore, the existence of DoS attacks can be known at a very
early stage and according to simulation results, the approach uses memory efficiently. In [45], the same
authors present another scheme based on Bloom filter and random deterministic message marking (IP-
CHOCK) in order to detect DoS attacks. The mechanism should be deployed at edge routers in order to
take place near the possible attacker. The packets are marked when they arrive at the edge router. The real
source IP addresses of vehicles are included in the marking fields. Thus it is possible to reconstruct
addresses at the destination. There are three phases in the detection process. IP addresses are collected
during detection engine phase 1 and checked as to whether or not they are malicious IPs during detection
engine phase 2. In the Bloom-filter phase, if they are malicious, an alarm is raised and a reference link is
sent to other vehicles. Evaluation results show that the proposed scheme effectively detects the attacks.
However, both of the proposed schemes require a dedicated infrastructure.
Kerrache et al. [46] develop a framework called TFDD based on trust establishment between vehicles
which is able to detect DoS and DDoS attacks in a distributed manner. Each vehicle V utilizes some
parameters by using its various modules. The first parameter is honesty weight (H) which is initially
assigned as 1 to each neighbor N. If V receives more packets from neighbor N than the thresholds which
are predetermined, the Intrusion Detection Module punishes N by locally decreasing its H value. The
second parameter is quality weight (Q) which refers to overall quality of packets received from N. These

two parameters are periodically combined by the Delayed Verification Module in order to generate DoS
weight (DW) which will be used as another parameter to globally decide whether or not a DoS attack
exists. On the other hand, trust weight (TM) of every message M received from vehicle N is combined with
the Q parameter by the same module. If the TM value is below a predefined threshold, the minimum of the
two parameters is selected as the new TM value to impose a fine on N. After that, in the Decision Module,
TM and DW parameters are combined in order to calculate the updated trust value between V and N (TN).
If both TN and DW values are not between the two predefined thresholds, it is possible for vehicle V to
include its neighbor N to its local blacklist. In that case, N could be included to the global blacklist and
suspended from network operations by a trusted authority as proposed in [47]. Even though the scheme is
able to detect the attack effectively according to simulation results, determining the thresholds in an
adaptive manner could have increased detection rates.
DoS attacks can be implemented at different layers, hence different solutions are proposed for each layer.
Since a DoS attack usually refers to exhausting available resources, determining threshold values is the
most preferred method of the aforementioned solutions. Determining those thresholds dynamically is more
suitable for VANETs due to its highly mobile and dynamic environment. In addition, almost all solutions
presented have some kind of response mechanism to fight against an attack.
3.3. Blackhole Attack
Blackhole attacks result in packet losses in the network and it could affect life-or-death decisions where
safety-related applications cannot send or receive critical data. For instance, a vehicle which is involved in
a traffic accident should propagate warning messages, but an attacker could prevent others from receiving
the warning by misrouting packets. Solutions which address such an attack are as follows:
Hortelano et al. [48] evaluate the protocol independent watchdog mechanism [9] in VANETs. In
watchdog, if node A sends a packet to node B, node A can check whether or not node B forwards the
packet by covertly listening to node B’s transmissions due to the nature of wireless medium. In the
proposed solution, each vehicle uses a neighbor trust level for each neighboring vehicle. Neighbor trust
level can be determined as the ratio of packets sent to the neighbor and the packets which are actually
forwarded by the neighbor. However, packets may not be forwarded because of a collision, as well as due
to an attack. Hence, a tolerance threshold is determined in order to prevent the false accusation of benign
vehicles. Consequently, if a vehicle keeps on dropping packets until the tolerance threshold is exceeded, it
will be considered as malicious and an alert message is generated. However, as the authors mentioned, a
watchdog mechanism can be deceived when there are two consecutive attackers in the network.
Daeinabi et al. [49] propose an algorithm which utilizes vehicles monitoring each other in order to detect
vehicles explicitly drop or duplicate packets. The vehicles are grouped into clusters and the Cluster Head
(CH) is the most trustworthy vehicle in each cluster which can be selected dynamically. When a vehicle V

joins the cluster, verifiers of V start monitoring the behavior of V. Verifier vehicles have to be more
trustworthy than V and located in such a way that they are able to monitor V and report to the CH. If a
verifier observes that V is dropping or duplicating packets, it reports V to the CH. After that, the CH
increases the distrust value of V, and informs V’s neighbors about the new distrust value. If that value
becomes higher than a threshold, CH reports V to the Certificate Authority (CA). The CA adds V to the
main blacklist and informs all vehicles. As a result, other vehicles isolate V from the network by not
communicating with it. The simulation results show that the proposed algorithm could detect possible
attackers at high speeds. However, it causes high end-to-end delay and jitter [50]. The algorithm is
improved by including a prevention feature in [50] and by selecting verifiers more effectively in [51].
Wahab et al. [52] proposed a mechanism which also utilizes watchdog mechanism in order to detect
selfish behaviors as well as Blackhole attacks against QoS-OLSR [53]. It is a five-phase detection
technique based on Tit-for-Tat concept and Dempster-Shafer theory of evidence. In the reputation
calculation phase, initial reputation values are assigned to vehicles and Multi-Point Relay (MPR) nodes are
selected by elected cluster-heads in order to forward traffic to other clusters. In the watchdogs monitoring
phase, MPR nodes are observed by cluster members, as the name implies. Then, the cluster-head utilizes a
voting mechanism and aggregates observations to make a final decision for cooperation by using the
Dempster-Shafer theory of evidence during the votes aggregation phase. After that, in Tit-for-TaT
cooperation regulation phase, if the trustworthiness of a MPR is higher than a predetermined threshold, the
decision will be made to cooperate with the MPR. Finally, during the information dissemination phase,
cluster-heads broadcast the results to cluster members and other cluster-heads. As a result, the members
isolate the vehicle(s) which are classified as malicious. Simulations show that the proposed cooperative
mechanism performs better than techniques based on one-to-one watchdog decisions.
By using watchdog mechanism in a cooperative manner, Baiad et al. [54] proposed a solution based on
monitoring in both network (VANET-OLSR protocol) and MAC layers in order to detect blackhole attack
targeting the Multi Point Relays (MPRs). The solution utilizes the cooperative watchdog mechanism [55]
which relies on routing level monitoring and could falsely accuse benign nodes in cases of packet loss due
to legitimate collisions. In order to minimize the false positive rate, detections reported by the cooperative
watchdog are filtered by using MAC layer monitoring. If the number of the sent RTS packets is different to
the number of received CTS packets, the cause of the packet loss is a legitimate collision. The proposed
cross-layer scheme correlates the results from both layers and, after eliminating watchdog detections
caused by collisions, the attackers can be identified. Simulation results show that the scheme has a lower
false positive rate and a higher detection percentage than the cooperative watchdog mechanism. However,
randomly selected watchdog nodes, which are assumed as trusted, might be malicious and could affect the
performance of the proposed solution. In [56], the authors enhanced their cross-layer solution by

aggregating physical layer detection with MAC and network layer detection and improved the detection
rate.
Unlike other solutions based on watchdog mechanism, Guo et al. [57] introduced a detection mechanism
based on encounter tickets (ER), which is introduced in [58]. It is assumed that by using data about
previous encounters of a vehicle, it is possible to evaluate its behavior. After two encountering entities such
as vehicles and/or RSUs to successfully exchange data, they send a digitally signed ER which contains
information about the encounter to each other. The ER includes timestamp, IDs of both vehicles, and the
sequence numbers which are unique for each contact. Therefore, by exchanging ERs during an encounter, a
vehicle could provide intact information to others about its behavior. It is also assumed that each vehicle
has a Trust Reputation (TR) value which is decreased in case the vehicle V selectively drops packets. In
addition, each vehicle has a blacklist and a meeting list which contains information about all previously
encountered vehicles. Hence, if V has a TR value lower than a threshold or tries to forge or replay ERs,
other nodes could detect and isolate vehicle V by adding it to their blacklists. The authors also introduced
an adaptive threshold mechanism [59] and a more flexible approach for dense and sparse networks by using
cluster analysis in [60]. However, the attacker could still drop packets after gaining reputation by tailgating
[61].
Yao et al. [62], recently proposed an entity-centric trust model which could detect blackhole attacks.
Each vehicle uses three trust parameters initialized with default values. Firstly, each vehicle A calculates its
direct trust value to vehicle B based on B’s data forwarding rate and the forwarded data’s weight according
to its data classes: traffic safety, traffic efficiency, and infotainment data. Secondly, the recommendation
value of A to B is determined by utilizing A’s direct trust value to its neighbors and their comprehensive
trust values to B. Comprehensive trust, the third parameter, refers to a dynamically varying combination of
direct trust and recommendation parameters. If A’s calculated comprehensive trust value to B is lower than
the predefined threshold, B could be detected as an attacker and isolated. Even though the proposed model
could detect a blackhole attack while maintaining the routing performance at a reasonable level according
to simulations, the rate of malicious vehicles in the network is not specified in the simulations.
To conclude, watchdog mechanism, which refers to checking the forwarding state of the forwarded
packets by monitoring the next hop neighbor over wireless media, is the most chosen method to ensure that
a neighbor is not a blackhole attacker. As shown in [55], utilizing watchdog mechanism in a cooperative
manner could increase the detection rate.
3.4. Wormhole Attack
In a wormhole attack, the attackers use tunnel(s) which attract a large amount of network traffic.
Therefore, collecting critical network data becomes possible for the attacker. Moreover, the attacker could
manipulate traffic and/or perform more aggressive attacks by analyzing collected data. Attack detection is

difficult since it does not generally affect normal network operations. Proposed works for detection are
presented as follows.
Safi et al. [63] introduced a solution which relies on leashes for on-demand routing protocols. A leash is
a mechanism that controls the packet’s maximum and allowed transmission distance by using information
in the packet. The authors prefer geographical leashes in order to guarantee that the distance between the
sender and receiver of a packet is no larger than a certain value. It is assumed that all vehicles know their
own location and that they have loosely synchronized clocks. When a vehicle wants to send a packet, the
sending vehicle includes the time and its own location in the packet. The receiving vehicle then compares
these values to its own parameters. By multiplying hop count value and predefined maximum transmission
range value for a vehicle, it is possible to calculate the maximum travel distance (x) of a received packet. If
x is smaller than the physical distance between the sender and receiver, or the packet has moved faster than
a predefined maximum velocity threshold, the attack is detected. In addition, packets must be authenticated
at each hop by using a HMAC-based algorithm in order to prevent modification and provide non-
repudiation. In a similar manner, if vehicles’ location parameters are involved in message authentication as
in Biswas et al. [64], it is possible to detect the attack by identifying tunneled packets which have been
transmitted over maximum allowed distances.
3.5. Bogus Information Attack
Disseminating bogus information could cause different results with respect to the intent. For example, a
selfish attacker that manipulates other vehicles to take alternative roads is generally considered harmless.
However, a malicious attacker could do the same in order to cause serious accidents. We researched and
categorized previous works about disseminating false information and the results are presented as follows.
The proposed solutions use a wide range of mechanisms such as verifiable multilateration and watchdog,
since there are numerous kinds of false information which could be disseminated by the attacker.
Kim et al. [65] propose a message filtering model to effectively detect bogus information. The model
includes a threshold curve and a certainty of event (CoE) curve. The CoE, which indicates the confidence
level of a received message, is calculated by combining the data from various sources such as local sensors,
RSUs and reputation mechanism. Priorities of the sources could be changed by the specific application for
the event so as to minimize computation. For example, a warning message about the road condition that
comes from a nearby trusted RSU could override other sources and make it unnecessary to check them. The
more a vehicle approaches to a real event, the more it receives messages reporting the event which
increases the CoE value. Since the solution relies on honest majority, an attacker which controls a small
fraction of the network cannot deceive the vehicle. The threshold curve shows the insensitivity of the driver
with respect to the distance to the event. Sensitivity and the distance to the event are inversely proportional.
Therefore, while the threshold value is decreasing, the CoE value keeps increasing and, if it exceeds the

threshold value which is assigned according to the application, the driver is warned with an alert message
and the reputations of vehicles which reported the event will be increased. Otherwise, the alert is discarded;
meaning it is considered bogus and the reputations of corresponding vehicles will be decreased and if they
continue lying, their messages may be filtered. However, on the whole, it does not address bogus
information attacks in various applications except the Electronic Emergency Brake Light (EEBL)
application, which enables broadcasting self-generated emergency brake event information to nearby
vehicles.
Another solution called Misbehavior Detection System (MDS) was devised by Raya et al. [66] to detect
neighboring attackers disseminating bogus information. It is assumed that there is a Certificate Authority
(CA) which provides one public/private key pair and a certificate for each vehicle. Also they use entropy, a
typical measure of information, to make it possible for a vehicle to represent and compare normal and
anomalous behaviors in order to detect the attacker. After that, the vehicle performs k-means clustering,
partitioning observation space into k clusters in a way that the sum of distances of all points to the
corresponding cluster centroids are minimized. Therefore, the vehicle could detect which neighbor(s)
differentiates from other neighbors since it is the attacker. For instance, if high velocity information
received from a neighboring attacker is contradictory to the state of the honest majority in a traffic jam, it
will be detected. After detection, another proposed protocol, Local Eviction of Attackers by Voting
Evaluators (LEAVE), is activated to isolate the attacker by revoking its certificate. Although the system is
effective and efficient, it requires an honest majority. If there are too many compromised neighbors, they
could dominate what “normal” behavior looks like and the attacker could remain undetected.
Ghosh et al. [67] propose a scheme for post-crash notification (PCN) applications. It is assumed that the
trajectories which a vehicle V follows in case of no alert and receiving a PCN alert conforms a free flow
mobility model and a crash-modulated mobility model, respectively. Upon receiving a PCN alert, V
analyzes its driver’s behaviors in response to the alert for a while and compares the actual trajectory and the
expected trajectory based on the model. Based on the deviation, the alert is considered as valid or not. For
example, if V comes across a crash site and the location of the site is different from the one reported by the
PCN alert, V’s actual trajectories deviate from what was expected. Therefore, it is possible for each
individual vehicle to identify the root-cause of misbehaviors by utilizing a cause-tree. After that, the
scheme will be able to determine a response with respect to the identified root-cause. Even though the
proposed scheme provides reasonable detection rates according to simulations, modeling expected
trajectories for every possible alert situation in a sensitive manner is not a feasible solution [5].
3.5.1 False Position Information
In VANETs, false position information could cause a severe decrease in the overall packet delivery ratio
which affects reliability. Since safety-related applications need reliable position information, serious safety
issues could arise. Various approaches are proposed for detection.

Vora et al. [68] devise a solution for position verification in a region R. The solution includes two kinds
of verifiers: acceptors and rejecters. While acceptors are distributed over the region R, rejecters are placed
around the acceptors in a closed circular fashion. It is assumed that secure and reliable communication
between verifiers is provided and verifiers are synchronized. When a node sends a signal, position
verification is performed by the verifier which receives it first. If the signal is first received by an acceptor,
it is verified that the node is located within region R. Despite the fact that the given approach is versatile, a
node can spoof its location as soon as it resides within region R since it does not verify the exact location of
the node. Although developed for mobile devices, it can also be implemented in vehicular networks. In
VANETs, the approach can be implemented with multiple RSUs surrounding one RSU in order to check if
a vehicle’s claimed position is false or not.
Fig. 6. Verifiable multilateration
By utilizing similar physical characteristics, Hubaux et al. [69] present verifiable multilateration which
relies on distance bounding [70] for position verification in VANETs. Distance bounding is a technique to
estimate physical distance. It is based on the fact that light travels at a finite speed. Therefore, by measuring
the time between sending a message to vehicle A and receiving its corresponding message, it is possible to
determine an upper bound on physical distance between RSU and vehicle A. An attacker vehicle could
show itself further away by delaying messages, but will be detected if it tries showing itself closer to the
RSU. Therefore, in a case there are two RSUs, vehicle A could not deceive both of them. If the same
technique is employed by more than one station in different dimensions, it is called verifiable
multilateration which enables estimating the exact location of vehicle A. In Fig. 6, the attacker vehicle can
claim it is at location L by delaying messages coming from base station B1, however, it is impossible to
reply to messages coming from B2 and B3 as if it is located at position L. The authors show that three
stations positioned to form a triangle can verify vehicle A’s location in two dimensions if it resides within
the triangle and four verifiers forming a triangular pyramid can do the same operation in three dimensions.
Although the verification process is fast, the solution requires a dedicated infrastructure and according to
B3
B1
B2
L
: Attacker
L : Fake Position

[71], an attacker can send delayed responses to each station by using directed antennas to deceive the
system.
Leinmüller et al. [71] propose another position verification approach based on a trust model. It is
assumed that there are two groups of sensors which provide observations. The first group works
autonomously and provides the necessary information in order to detect false (invalid or incorrect) position
information. As an example of this, the Acceptance Range Threshold (ART) sensor discards packets which
include position information that is further than the predefined maximum acceptance range threshold.
Obtained information is weighted according to the reliability of the providing sensor and used to determine
the overall trust ratings of neighbors. For example, information provided by an ART sensor is more reliable
than the information provided by a Map-Based Verification mechanism which checks whether or not a
vehicle is at an invalid location (e.g. off road). Because of weighting abnormal observations that are higher
than normal, a vehicle’s trust level is more affected by abnormal observations. Therefore, after sending one
bogus information packet, the vehicle has to send a few correct information packets in order to regain its
previous trust level. The second group works cooperatively with other vehicles. For instance, Proactive
Exchange of Neighbor Tables mechanism lets the vehicles exchange and compare neighbor tables so as to
believe the majority in case of false information. The approach needs no infrastructure and, according to
the simulation, it detects vehicles which disseminate false positions with an accuracy of 95% in most
scenarios. However, an attacker vehicle can still disseminate bogus information as long as it meets the
criteria being enforced by the sensors. Moreover, it can deceive its own sensors, as described in the
subsequent section.
Bissmeyer et al. [72] propose a signature-based scheme which relies on a plausibility model. Each
vehicle is modeled as differently sized and nested rectangles, and intersecting rectangles that belong to
different vehicles could indicate false position information. Due to inaccuracies of positioning systems such
as GPS, the probability of a real intersection is associated with intrusion certainty and trust values by each
individual vehicle. Intrusion certainty value is calculated based on the number of observed intersections,
and the attack is not identified unless it is above a predetermined threshold. In addition, by utilizing the
Minimum-Distance-Moved (MDM) concept proposed in [73], the trust values are also used for detection in
such a way that when vehicle N remains as a neighbor of V for a distance which is larger than the
transmission range, N becomes more trustworthy as a result. Therefore, from V’s point of view, when N
intersects with another neighbor and the difference between trust levels of both vehicles is higher than a
threshold, the less trustworthy vehicle is identified as an attacker. According to the simulation results, the
proposed plausibility model could detect false position information in spite of GPS error(s). However, an
attacker whose transmission range is larger than other vehicles could bypass the trust mechanism.
Ruj et al. [74] devise a data-centric misbehavior detection system (MDS) which can be used to detect
false location information. The idea of data-centric MDS concept is to classify data instead of classifying

vehicles. Each vehicle can verify the location information independently by using the proposed technique.
An attacker sends a beacon message which includes fake location information (L1) along with the
timestamp it is sent (T1). After it is received by a vehicle V located at L2 at time T2, V can verify if L1 is
correct or false by utilizing L1, L2, speed of light and the difference between T1 and T2. In that scenario,
the attacker is not able to modify T1 in order to deceive V since it does not know the exact distance
between itself and V. If the location is detected as false, V broadcasts a message to other vehicles and the
CA through the nearest RSU. This leads to fines imposed on attackers instead of isolating them from the
network. The authors compared the proposed scheme with existing MDSs in terms of communication
overhead; however, it is not supported with simulations.
3.5.2 Sensor Tampering
Sensor tampering can be easily performed. All the attacker needs to do is to trick its own sensor(s) into
reporting valid, but false messages. For example, the attacker can use some ice in order to deceive
ice/temperature sensor(s) and generate messages indicating false road conditions. Therefore, it is possible
to disseminate false information to the network even though message integrity remains protected.
3.5.2.1 Illusion Attack
In an illusion attack, the attacker needs a scene which enables the deceiving of other drivers easier, in
addition to tricking sensors. Therefore, it is possible to influence other drivers’ behaviors by disseminating
bogus information which is supported by traffic situations.
Lo et al. [24] introduce the illusion attack and a plausibility mechanism called Plausibility Validation
Network (PVN), which is capable of checking the output data of sensors and deciding whether or not the
data is valid. The proposed solution includes a plausibility network (PN) checking module and a rule
database. The system for any message which is received from other vehicles or generated from the sensors
works as follows: Each message is evaluated with respect to its type (accident, general road condition, etc.).
A corresponding predefined rule set is retrieved from the rule database for a PVN in order to check
reasonability of the value of an element field (timestamp, velocity, etc.) of the given message. For example,
similar to other fields, the plausibility of the timestamp field is checked by determining its minimum and
maximum boundaries. The received timestamp value must be earlier than the receiver’s current timestamp
and later than the difference between the receiver’s current timestamp and the validity period of the
message. Lastly, the values are cross-verified with values of other relevant element fields. If the
verification of all element fields in the message is completed successfully, the message will be considered
as trustworthy. Otherwise, it will be discarded.

3.5.2.2 GPS Spoofing
Attackers could convince victim(s) to think that they are in a different location. To this end, they
generally use GPS simulators which generate more powerful signals than the original GPS generates.
Studer et al. [75] propose a simple mechanism to detect GPS spoofing by dead reckoning. Dead reckoning
calculates one’s current position by using a previously determined position and known or estimated speeds
over elapsed time. Although it can detect and filter out spoofed GPS information, the calculated position is
only approximate.
3.6. Replay Attack
In replay attack, legitimate data which is gathered or originated by the attacker is simply stored for use
later. Therefore, as long as the data remains valid, the attacker can get the same results as recurring the
original situation. A method which utilizes traditional timestamp mechanism for detecting the replay attack
[76], could be employed for VANETs. A node that receives a message checks the timestamps. If the
difference between current and received timestamps is larger than a predefined threshold, the message is
rejected or dropped. The key point of detection of the attack is ensuring timestamp integrity.
3.7. Solutions for Multiple Misbehavior Detection
Yan et al. [77], devise a solution based on “seeing is believing” which enables fast detection of
misbehaving vehicles. They assume that each vehicle has a front and a rear short-distance radar which are
used as the virtual “eye” of the vehicle. By using radar detections, neighbors’ data and incoming traffic
data provided by vehicles which are further than one hop, it is possible for a vehicle to “see” whether
received data are fake or not. After that, similarity of data provided by those sources is calculated. Different
sources have different weights. While radar detections have the highest weight due to their high level
reliability, incoming traffic data has the lowest weight. If the similarity among the data collected from three
different sources is close to each other, the data will be accepted. Otherwise it will be discarded. The
accepted data will be used to compute the average position and velocity. After that, map history including
the vehicle’s past observed movements will be built. By checking map history, it will be possible to
determine whether the given position(s) are consistent or not. The solution requires line-of-sight (LOS)
between vehicles. However, even when LOS is limited temporarily, it is still possible to detect an attacker
by correlating information coming from “eyes” and received from other vehicles. According to the
simulation results, the approach is efficient in detecting malicious vehicles. In [78], the authors extended
their solution by applying a similarity model and adding more “eyes” such as infrared detectors and CCD
(charge coupled device)-based sensors which can cooperate with each other. However, a Sybil attack could
still be performed while LOS between vehicles is blocked.
Kumar and Chilamkurti [79] develop an adaptive scheme called T-CLAIDS, which is based on Learning
Automata (LA) and watchdog mechanism. Each node runs a particular code called automaton which takes

the density, mobility and direction of motion of the vehicles as inputs from the environment in order to
perform actions such as data collection, detection and alert generation to produce outputs. The automaton
also calculates a Collaborative Trust Index (CTI) and updates its value with respect to received rewards and
penalties from the environment which monitors performed actions. If the CTI value is lower than a
predefined threshold, a malicious activity is detected in that region and generated alarms are immediately
transferred to other vehicles since automatons share information with each other. According to performance
evaluation results, the solution is capable of detecting malicious activities and is scalable.
Another watchdog-based solution called AECFV, which is a framework based on clustering and utilizing
RSUs, is proposed by Sedjelmaci and Senouci [80]. AECFV includes three systems. While Local Intrusion
Detection System (LIDS) and Global Intrusion Detection System (GIDS) are for detection, Global Decision
System (GDS) is used to make final decisions. LIDS runs at member vehicles within a cluster and makes
monitoring their neighbors and the cluster-head behaviors possible in their transmission range. In order to
reduce overhead, vehicles do not activate their LIDS until they reach an optimal state called Nash
equilibrium, a game theory concept. GIDS runs at each cluster-head and allows it to evaluate
trustworthiness of cluster member vehicles by monitoring them. GDS runs at each RSU and categorizes
vehicles into lists such as blacklist or suspected list by calculating their trust level. The framework can be
used to address blackhole, wormhole, selective forwarding and Sybil attacks by utilizing watchdog
mechanism and calculating adaptive thresholds for send/dropped packet ratios and signal strength intensity
with the help of learning algorithms based on Support Vector Machine (SVM). Simulation results show
that AECFV effectively detects such attacks.
The watchdog mechanism was also used by Wahab et al. [81] to develop an intelligent detection model
called CEAP which is based on SVM and Quality of Service Optimized Link State Routing protocol
(VANET QoS-OLSR) in order to detect misbehaving MPRs. The model includes four phases: during the
data collection phase, cluster head and other cluster members monitor the behaviors of MPR nodes in a
continuous manner. After the data exchange phase, which includes sharing collected information with
other members of the cluster, each monitoring vehicle utilizes SVM to classify MPRs as cooperative or
malicious by using its own observations as test data and others’ observations as training data in the data
analysis phase. Finally, a data propagation phase is proposed to enable cluster heads to exchange classes
of MPRs. Even though simulation results show that the model has higher packet delivery ratio and
detection rate than existing detection systems, it assumed the cluster head is a trusted third party.
Grover et al. [82] also proposed a machine learning approach in order to detect misbehaviors in
VANETs. Various forms of misbehaviors and legitimate instances are used to extract features. In order to
identify different types of misbehaviors, different features are extracted by using three inputs, which are the
proposed VANET model based on some assumptions such that RSUs are always honest, the attack model,
and the VANET application affected by the attack. After conducting experiments with different

combinations of attacks, the extracted features calculated by observer nodes are as follows: speed
deviation, distance, received signal strength (RSS), and the number of generated/delivered/dropped/collided
packets. In order to classify the behavior of a vehicle as honest or malicious, the authors used Naive Bayes,
IBK, J-48, Random Forest (RF), and AdaBoost1 classifiers and compared their results. The comparison
shows that RF and J-48 classifiers perform the best. However, as the authors stated, the proposed solution
may not be suitable in detecting temporal attacks such as replay attacks in a realistic VANET scenario. In
order to increase detection performance, the authors improved their scheme by combining the results of
previously mentioned classifiers to reach a final decision using ensemble-based machine learning in [83].
Bouali et al. [84] developed Intrusion Prevention and Detection System (IPDS), a predictive approach
which is able to detect multiple misbehaviors before they can take place by predicting vehicles’ future
behaviors. The vehicles are grouped into one-hop clusters and there are three roles within each cluster.
Firstly, the most trustworthy vehicle in each cluster is elected as CH. Then, three recommenders are
selected by the CH in such a way that the CH divides its communication range into three equal regions and
the most trusted vehicle closest to each region center is selected as a recommender. After that, the rest of
the vehicles become member vehicles which are monitored by the recommenders. The CH permanently
monitors the member vehicles and updates their trust levels by using its own observations and the
information received from the recommenders. In order to detect a future attack, the CH utilizes Kalman
filtering, a method that is able to predict future trust levels by using current trust levels and the previously
predicted information as inputs. In addition, according to prediction results, the CH classifies the vehicles
into white (benign), black (malicious), and gray lists. The gray list includes vehicles which intermittently
misbehave. Moreover, the CH is responsible for broadcasting the identities of the vehicles in black and
gray lists. Therefore, vehicles in the blacklist could be isolated from the rest of the network and vehicles in
the gray list could be used for routing purposes when there are no vehicles in the whitelist. Also they can be
moved to whitelist or blacklist depending on their behaviors. According to simulation results, IPDS is able
to detect various misbehaviors effectively and accurately. However, if multiple attackers behave normally
until seizing the CH and recommenders roles, an attack may take place.
Kerrache et al. [85] developed a trust architecture called T-VNets which mainly evaluates two trust
parameters in parallel: inter-vehicles trust and RSUs-to-vehicles trust. Inter-vehicles trust is built by
combining data centric evaluation of messages received from each neighbor vehicle N with received
reports about the neighbor. These reports are broadcast by the neighbors of N who detect a positive or
negative change in the behavior of N by using a watchdog mechanism. RSUs-to-vehicles trust means that
RSUs collect reports from vehicles, about their neighbors’ behaviors which help them to have a quasi-
global trust value of each vehicle historically and regionally, since RSUs are usually connected to each
other. After that, inter-vehicles and RSUs-to-vehicles trust are combined to estimate a global trust value for
each vehicle. The estimation could also be performed in a limited manner by each vehicle if there is no

RSU within its vicinity. Then, global trust values will be inserted to periodically exchanged Cooperative
Awareness Messages (CAM) through the addition of new fields. CAM is an ETSI (European
Telecommunications Standards Institute) ITS standard which enables the periodic exchange of information,
dynamics and attributes between vehicles and infrastructure [86]. While CAMs are used to evaluate
regional trust, another ETSI ITS standard called Decentralized Environmental Notification Messages
(DENM) are used to dynamically calculate a specific event’s trust since DENMs are triggered based on
road hazards [87]. The global trust value of a vehicle V reporting the event and the data centric evaluation
of DENM message are used to calculate the event’s trust. Therefore, events which have a lower trust value
than a predefined threshold will not be broadcast by an intermediary vehicles and the V’s global trust level
will be updated. As a whole, by determining threshold limits for calculated trust values, it is possible for a
vehicle to effectively detect a malicious vehicle.
Last but not least, Zaidi et al. [88], developed a scheme which relies upon statistical techniques in order
to detect multiple misbehaviors. It utilizes a model named Greenshield’s Model to predict and explain the
trends in real traffic flows. Each vehicle could estimate its own flow parameter (F) which should be very
similar for vehicles located closely to each other by using a model that employs density referring to
vehicles per kilometer and the average speed of other vehicles in its vicinity. After that, vehicles exchange
their own F and density values along with their speed and location information; hence each vehicle could
have information about surrounding vehicles. For each received message, vehicles compare the average of
the received parameters to its own calculated parameters. If the difference is lower than a threshold, the
message is accepted. Otherwise, the sender is monitored and the data is accepted until the number of
collected messages is enough to perform a t-test. Results of the test determine whether or not the sender is
malicious. Then, the malicious vehicle will be reported to other vehicles and isolated from the network by
rejecting its data. According to simulations, the proposed scheme could detect malicious nodes even when
40% of the vehicles are malicious. However, an attack performed in a low and slow manner may remain
undetected by the system since the attacker only manipulates values gradually.
The solutions presented here mainly rely on watchdog mechanism to detect multiple misbehaviors. The
reason is that they are mostly developed to let vehicles perform intrusion detection independently. This
feature makes them deployable more easily and more quickly than other solutions since there is no need for
a dedicated infrastructure. Furthermore, it is shown that artificial intelligence-based approaches with
watchdog mechanism proposed in recent years have a potential in detecting multiple misbehaviors.
4. DISCUSSION
VANETs are continuously developing and attracting increased attention. Consequently, new attacks are
being discovered, with impacts ranging from misrouting vehicles to traffic accidents. Attacker motivation
plays a critical role on the effect of an attack. For example, a driver could be deceived by an attacker to

decelerate. This attacker could be a person who just wants to mess with other drivers for fun, or a terrorist
who tries to create traffic congestion before a bombing attack. The attack surface even enlarges with the
existence of IoV. In that case, an attacker could compromise some vehicles and turn them into zombie
vehicles, awaiting orders from a command and control (C&C) server.
It is also significant to note that the number of road traffic deaths worldwide was 1.25 million in 2013
[89]. Safety related applications in VANETs such as post-crash notification and road hazard notification are
expected to lower that number in the near future. However, they require real-time network operations
which also demand message encryption at the same time. Therefore, for safety-related applications,
intrusion prevention has higher priority than intrusion detection. Nevertheless, in the case of a highly
motivated attacker who specifically targeted those applications and already bypassed the existing
prevention systems, intrusion detection must also be performed in real-time. Otherwise, there could be
catastrophic consequences such as traffic accidents and delayed rescue operations. In this section, first an
outline is provided on the proposed approaches from the intrusion/misbehavior detection perspective, then
other proactive and reactive solutions that could be employed as countermeasures to attacks are discussed.
Finally, open issues are presented as directions to be considered for future research.

Table 1 – Outline of the Proposed Approaches
#
Study
Infrastructure
Proposed Method
Reputation
Mechanism
Response
Mechanism
Attack(s) Covered
1
Golle et al. [29]
OBU-based (standalone)
Comparing received data to the model
No
Correction
Sybil Attack
2
Xiao et al. [30]
Hybrid
Signal strength analysis
No
-
Sybil Attack
3
Zhou et al. [33]
RSU-based
Hashing pseudonyms to common values
No
Revocation
Sybil Attack
4
Rahbari et al. [38]
RSU-based
Public key Infrastructure
No
Revocation
Sybil Attack
5
Yong Hao et al. [36]
OBU-based (cooperative)
Watchdog mechanism
Yes
Isolation
Sybil Attack
6
Lee et al. [37]
RSU-based
Session-key-based certificate
No
-
Sybil Attack
7
Chen et al. [41]
Hybrid
Observing similarity of motion trajectories
No
-
Sybil Attack
8
Park et al. [34]
Hybrid
Observing similarity of motion trajectories
No
-
Sybil Attack
9
Feng et al. [39]
Hybrid
Public key Infrastructure
Yes
-
Sybil Attack
10
Grover et al. [42]
OBU-based (cooperative)
Observing similarity of motion trajectories
No
-
Sybil Attack
11
Chang et al. [40]
Hybrid
Observing similarity of motion trajectories
No
-
Sybil Attack
12
Soryal et al. [43]
RSU-based
Markov Chain model
No
Isolation
DoS Attack
13
Verma et al. [44]
RSU-based
Monitoring outstanding SYN packets
No
-
DoS / Flooding Attack
14
Verma et al. [45]
RSU-based
Packet marking
No
Alarm
DoS Attack
15
Kerrache et al. [46]
OBU-based (cooperative)
Trust model using transmission thresholds
Yes
Isolation
DoS Attack
16
Hortelano et al. [48]
OBU-based (standalone)
Watchdog mechanism
No
-
Blackhole Attack
17
Daeinabi et al. [49]
Hybrid
Watchdog mechanism
Yes
Isolation
Blackhole Attack
18
Guo et al. [57]
Hybrid
Encounter tickets
Yes
Isolation
Blackhole Attack
19
Wahab et al. [52]
OBU-based (hierarchical)
Watchdog mechanism
Yes
Isolation
Blackhole Attack
20
Baiad et al. [54]
OBU-based (hierarchical)
Watchdog mechanism
Yes
-
Blackhole Attack
21
Yao et al. [62]
OBU-based (cooperative)
Trust model based on weights
Yes
Isolation
Blackhole Attack
22
Safi et al. [63]
OBU-based (standalone)
Packet leashes
No
-
Wormhole Attack
23
Kim et al. [65]
Hybrid
Message filtering model
Yes
-
Bogus Information Attack
24
Raya et al. [66]
OBU-based (cooperative)
Entropy and k-means clustering
No
Revocation
Bogus Information Attack
25
Ghosh et al. [67]
OBU-based (standalone)
Observing deviation of motion trajectories
No
-
Bogus Information Attack
26
Vora et al. [68]
RSU-based
Time Difference
No
-
Bogus Information Attack
27
Hubaux et al. [69]
RSU-based
Verifiable multilateration
No
-
Bogus Information Attack
28
Leinmüller et al. [23]
OBU-based (cooperative)
Trust model using sensors
Yes
-
Bogus Information Attack
29
Bissmeyer et al. [72]
OBU-based (cooperative)
Plausibility model vehicle movements
Yes
-
Bogus Information Attack
30
Ruj et al. [74]
Hybrid
Watchdog mechanism
No
Fine Imposition
Bogus Information Attack
31
Lo et al. [24]
OBU-based (standalone)
Plausibility validation network model
No
-
Illusion Attack
32
Studer et al. [75]
OBU-based (standalone)
Dead reckoning
No
-
GPS Spoofing
33
Adjih et al. [76]
OBU-based (standalone)
Timestamps
No
Packet Dropping
Replay Attack
34
Yan et al. [77]
OBU-based (cooperative)
Computing similarity among collected data
Yes
Isolation
Multiple Misbehaviors
35
Grover et al. [82]
Hybrid
Machine Learning
Yes
-
Multiple Misbehaviors
36
Kumar et al. [79]
OBU-based (cooperative)
Watchdog mechanism
Yes
-
Multiple Misbehaviors
37
Sedjelmaci et al. [80]
Hybrid
Watchdog mechanism, Machine Learning
Yes
-
Multiple Misbehaviors
38
Wahab et al. [81]
OBU-based (hierarchical)
Watchdog mechanism, Machine Learning
No
Isolation
Multiple Misbehaviors
39
Bouali et al. [84]
OBU-based (hierarchical)
Watchdog mechanism
Yes
Isolation
Multiple Misbehaviors
40
Zaidi et al. [88]
OBU-based (cooperative)
Statistical data analysis
No
Isolation
Multiple Misbehaviors
41
Kerrache et al. [85]
OBU-based (cooperative)
Trust model using watchdog mechanism
Yes
Isolation
Multiple Misbehaviors

4.1. Outline of Proposed Approaches
In this work, a compilation has been created of the attacks and detection mechanisms proposed in the
literature. Table 1 outlines the proposed solutions. Most of the solutions in Table 1 are OBU-based only
which means they do not need a dedicated infrastructure. Vehicles performing situation evaluation by
themselves without any infrastructure makes the detection process faster through lowering the detection
time. Furthermore, due to highly mobile vehicles moving in very large areas, if central administration
mechanisms were to be employed in intrusion detection by deploying RSUs that cover large areas, this
could create high cost overheads, especially in rural areas. Therefore, if cost is a primary concern, OBU-
based only approaches would be preferable to RSU-based only or even hybrid approaches. However, cost is
not the only consideration if confidentiality and non-repudiation are required in order to detect attacks.
OBU-based approaches are divided into three sub-architectures as standalone, cooperative, and
hierarchical. Most of the OBU-based solutions are collaborative even though there are few standalone
mechanisms such as [48][63] in which each node detects attacks on its own. Hierarchical architecture is
also distributed and cooperative in nature, differently it divides network into groups such as clusters, and
gives more responsibility to some nodes such as cluster heads. This architecture is generally preferred for
vehicular ad hoc networks using clustered routing protocols such as QoS-OLSR [55][56].
Similar to some attacks which are inherited from MANETs to VANETs, some detection mechanisms
such as [68] can be inherited in the same way. Therefore, we presented the mechanisms not only for
VANETs but also some other mechanisms which originally developed for MANETs that could be
applicable to VANETs. However, inheritance from MANET to VANET does not always work well
because of VANETs’ special characteristics such as high mobility. On the other hand, enhancing a
vehicular ad hoc network with infrastructure could help vehicles in the detection of attacks as opposed to
MANETs that are lack of central points, but it is a costly approach due to deploying RSUs on a large scale.
Furthermore, many solutions proposed for MANETs are specific to the routing protocols used. However,
mainly protocol-independent solutions are proposed for intrusion detection in VANETs. Only some
approaches [52][81] use a particular architecture built as a result of routing protocol for detection. Since
solutions are mainly proposed for specific protocols in MANETs, specification-based detection, where the
violations of the set of constraints of a protocol defined is detected as attacks, is one of the most commonly
proposed techniques. On the other hand, to our knowledge there is no solution based on specification-based
detection for VANETs, anomaly-based intrusion detection, where intrusions are detected as deviations
from the normal behaviors, is generally employed in the solutions. In order to define what “normal” is, the
existence of honest majority in the environment is generally assumed in detection. However, that
assumption could become a disadvantage especially in case of a Sybil attack because an attacker can affect

decision processes by using the Sybil nodes to constitute a majority. There are also few signature-based
approaches [46], or hybrid approaches [72].
In general, solutions to detect the attacks can be reviewed as follows. The approaches on Sybil attack are
generally based on similar motion trajectories, identity registration, PKI, signal strength and/or sensors
such as radars. For DoS attack detection, packet marking and thresholds which are determined statically or
dynamically are utilized. Additionally, response mechanisms are developed due to the nature of the attack.
In order to detect a blackhole attack, redundant paths or watchdog mechanisms are utilized, while assuming
the existence of honest majority for both. Redundant paths are also used for wormhole attack detection.
When it comes to bogus information attack, the solutions mostly rely on sensors and/or challenge-response
procedures such as distance bounding to estimate distances. Moreover, location and digitally signed
timestamp information are also utilized in order to determine location difference. For the rest, the solutions
mainly based on machine learning and watchdog mechanisms are employed in order to detect multiple
misbehaviors. Although computational intelligence has many promising applications to intrusion detection
[90], there were only a few applications to VANETs [81][82] found in the literature.
While some approaches have response mechanism to detected attacks, they are mainly passive responses
such as raising alarms, or active attacks that seek control over the attacked system [91] such as isolation of
nodes and dropping malicious packets. Therefore, not only detecting attacks, but also identifying attackers
is important in order to properly respond to attacks.
The proposed solutions address different types of attack since they utilize various methods specific to the
attack(s). In the future, more protocol/application-specific attacks might need investigation. Please note that
this paper’s analysis is based on attacks taken into consideration for the purpose of detecting, as found in
the literature. It is the authors’ belief that each of the proposed solutions is a valuable contribution towards
securing VANETs. They are useful in order to build one or more layers of strong, standard security
architecture for VANETs, which could become the largest network in the future. It is hoped that this study
will also help VANET researchers and designers to develop more secure architecture in order to provide a
better transportation experience.
4.2. Other Countermeasures: Trust and Privacy
PKI, one of the most popular technologies to prevent intrusions in today’s networks, is offered in some
solutions such as [38] to provide encryption and authentication in VANETs. Digital signature is also used
to ensure integrity of the messages and non-repudiation. Since most of the attacks are performed in the
network layer, PKI and/or digital signatures are also used for routing operations. However, utilizing PKI
requires revoking digital certificates and distributing large CRLs (Certificate Revocation List) which is a
hot topic and suffers from real-time constraints. Moreover, using digital certificates cause concerns about

users’ privacy because users want to protect their private information like license plate and position. This is
a challenging task especially in detection of a Sybil attack because trusted certification is required in order
to guarantee that each entity has only a single identity. As a result, there is a trade-off between privacy and
non-repudiation. Apart from some solutions such as [33], most of the Sybil attack detection mechanisms
ignore privacy.
Trust is another key concept in vehicular networks as in human relationships. The concept brings
predicting the future to mind and it is built by gaining reputation which refers to knowledge of the past.
Reputation and cryptography are complementary mechanisms since the former could be used to detect
insider threats which could not be detected by the latter. Some of the proposed detection systems employ
reputation mechanisms in order to predict vehicles’ future actions based on their past behaviors. Therefore,
vehicles that deviate from a system’s expectations could be detected and isolated. After that, the network
could rely more on other (trustworthy) vehicles to perform network operations. This is especially the case
for blackhole attacks. Since the attacker exploits the trust of others by dropping the packets which are
expected to be forwarded, reputation mechanisms are mostly used to detect this type of attack. For further
information on trust management, the readers could refer to the brief survey provided in [92].
4.3. Open Issues
Considering its huge benefits to attacker(s) such as exploiting honest majority assumption and possible
serious consequences to others, it is not surprising that Sybil attack is the most preferred attack addressed
by researchers, as can be seen from Table 1. Nevertheless, drawing the line between privacy and non-
repudiation is still needed in detecting a Sybil attack. Furthermore, while there are numerous works on
Sybil attack, some other attacks such as illusion attack, motorway attack that are peculiar to VANETs
require further research. How to detect attacks and attackers, especially cooperative attack(er)s also needs
to be investigated. Distributed and cooperation intrusion detection architectures are more suited to the
problem. Even though using RSUs could be a costly approach, their future deployment could increase
security.
Some of the proposed attacks are performed on the routing layer where the routing protocols run were
originally developed without security in mind. On the other hand, the proposed secure routing protocols
require a high level of overhead and fail to address the demand for real-time operations and privacy at the
same time. As a result, implementation of a routing protocol which takes security into consideration and
could make the detection process faster while preserving privacy is an unsolved problem that could be
subjected to further study. Nevertheless, monitoring/securing one layer might not be enough for intrusion
detection. Even though most of the presented solutions focus on detecting misbehaviors at a specific layer,
the problem requires cooperation between different layers. There were only few cross-layer solutions

[54][56] found in the literature. Therefore, a study that implements an intrusion detection approach which
incorporates the protocol stack and makes different communication types such as V2V and V2I possible at
the same time is missing.
After resolving the abovementioned problems in detection, having response mechanisms still remains as
an important necessity for the solutions. It is especially critical for detecting a DoS or DDoS attack since it
is almost impossible to respond to the attack once it has been performed. The response mechanisms mostly
aim to isolate the attacker from the network as soon as possible. The isolation could be performed by
dropping/denying messages from the attacker, revoking the attacker’s certificate and/or avoiding sending
messages to the attacker. However, not all the proposed solutions have response mechanisms. It is
considered that response mechanisms require more attention and that researchers should work on adaptive
response mechanisms in order to discourage malicious and/or selfish activities in the future.
Another area that requires focus is the lack of reliable links. Many detection techniques collect data from
nearby entities in order to detect misbehavior(s) and suffer from a lack of reliable links between entities due
to their high level of mobility. In addition, high vehicle density could cause some problems such as
broadcast storms and disrupt the links in the network [93]. Therefore, making links more reliable could
make detection mechanisms more effective. Developing an intrusion detection architecture which provides
better detection rates by utilizing complementary technologies such as wireless communication and VLC
(Visible Light Communication) [94] at the same time remains as yet an unstudied area.
As previously stated, although computational intelligence has many promising applications to intrusion
detection [90], there were only a few applications to VANETs found in the literature. It is also shown that
machine learning-based approaches have the potential to discover complex properties of MANETs
[95][96]. The suitability of these techniques to intrusion detection in VANETs could be explored in future
studies. Moreover, they could be used to adaptively determine thresholds, which is necessary for such
highly dynamic environments. Since most of the studies found in the literature determine thresholds
experimentally and do not evaluate it in other scenarios, it is considered an open research area.
5. CONCLUSION
Conventional security approaches which are suitable for wired networks or even some MANETs cannot
be directly applied to VANETs due to their very characteristics. In this paper, the aim was to provide a
holistic view to previous works on intrusion/misbehavior detection in VANETs. Firstly, a survey has
presented an overview of the attacks, together with their possible effects along with working principals.
The techniques used in the attacks showed that the attackers generally exploit network and application
layers operations. Then a survey was presented of solutions using different detection mechanisms proposed

in the literature. Each was given along with the attack(s) they can address, along with the advantages and
disadvantages. Lastly, a survey of solutions was presented in Table 1 with respect to the methods used,
infrastructure, intrusion detection architecture, reputation, and response mechanisms. As can be observed
from Table 1, most of the solutions prefer using only OBUs instead of requiring a dedicated RSU
infrastructure, and employ the watchdog mechanism. Many do not provide a response mechanism, and only
cover few attacks. All solutions are discussed, and open issues outlined for future research. In conclusion,
attack/misbehavior detection in VANETs is a complex and challenging topic and, protocol stack-wide and
adaptive detection techniques, computational intelligence-based approaches are promising areas that could
be explored in future studies as a means to make VANETs more secure.
REFERENCES
[1] J. Wan, J. Liu, Z. Shao, A. Vasilakos, M. Imran, and K. Zhou, “Mobile Crowd Sensing for Traffic
Prediction in Internet of Vehicles,” Sensors, vol. 16, no. 1, p. 88, Jan. 2016.
[2] J. T. Isaac, S. Zeadally, and J. S. Cámara, “Security attacks and solutions for vehicular ad hoc
networks,” IET Communications, vol. 4, no. 7, p. 894, 2010.
[3] S. Şen and J. A. Clark, “Intrusion Detection in Mobile Ad Hoc Networks,” in Guide to Wireless Ad
Hoc Networks, S. Misra, I. Woungang, and S. Chandra Misra, Eds. London: Springer London, 2009,
pp. 427–454.
[4] A. Kapadia, N. Triandopoulos, C. Cornelius, D. Peebles, and D. Kotz, “AnonySense: Opportunistic
and Privacy-Preserving Context Collection,” in Pervasive Computing, vol. 5013, J. Indulska, D. J.
Patterson, T. Rodden, and M. Ott, Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2008, pp.
280–297.
[5] D. Antolino Rivas, J. M. Barceló-Ordinas, M. Guerrero Zapata, and J. D. Morillo-Pozo, “Security on
VANETs: Privacy, misbehaving nodes, false information and secure data aggregation,” Journal of
Network and Computer Applications, vol. 34, no. 6, pp. 1942–1955, Nov. 2011.
[6] “World Vehicle Population Tops 1 Billion Units,” 2011. [Online]. Available:
http://wardsauto.com/news-analysis/world-vehicle-population-tops-1-billion-units. [Accessed: 08-Jan-
2017].
[7] “Gartner Says By 2020, a Quarter Billion Connected Vehicles Will Enable New In-Vehicle Services
and Automated Driving Capabilities.” [Online]. Available:
http://www.gartner.com/newsroom/id/2970017. [Accessed: 08-Jan-2017].
[8] R. G. Engoulou, M. Bellaïche, S. Pierre, and A. Quintero, “VANET security surveys,” Computer
Communications, vol. 44, pp. 1–13, May 2014.
[9] S. Marti, T. J. Giuli, K. Lai, and M. Baker, “Mitigating routing misbehavior in mobile ad hoc
networks,” in Proceedings of the 6th annual international conference on Mobile computing and
networking, 2000, pp. 255–265.
[10] B. Parno and A. Perrig, “Challenges in securing vehicular networks,” in Workshop on hot topics in
networks (HotNets-IV), 2005, pp. 1–6.
[11] F. Stajano and R. Anderson, “The Resurrecting Duckling: security issues for ubiquitous computing,”
Computer, vol. 35, no. 4, p. supl22-supl26, Apr. 2002.
[12] “Gartner Says the Internet of Things Installed Base Will Grow to 26 Billion Units By 2020.” [Online].
Available: http://www.gartner.com/newsroom/id/2636073. [Accessed: 08-Jan-2017].

[13] M. Gerla, E.-K. Lee, G. Pau, and U. Lee, “Internet of vehicles: From intelligent grid to autonomous
cars and vehicular clouds,” in 2014 IEEE World Forum on Internet of Things (WF-IoT), 2014, pp.
241–246.
[14] O. Vermesan et al., “Internet of things strategic research roadmap,” O. Vermesan, P. Friess, P.
Guillemin, S. Gusmeroli, H. Sundmaeker, A. Bassi, et al., Internet of Things: Global Technological
and Societal Trends, vol. 1, pp. 9–52, 2011.
[15] A. M. Vegni, M. Biagi, and R. Cusani, “Smart Vehicles, Technologies and Main Applications in
Vehicular Ad hoc Networks,” in Vehicular Technologies - Deployment and Applications, InTech,
2013.
[16] I. A. Sumra, I. Ahmad, H. Hasbullah, and others, “Classes of attacks in VANET,” in Electronics,
Communications and Photonics Conference (SIECPC), 2011 Saudi International, 2011, pp. 1–5.
[17] I. Aad, J.-P. Hubaux, and E. W. Knightly, “Denial of service resilience in ad hoc networks,” in
Proceedings of the 10th annual international conference on Mobile computing and networking, 2004,
pp. 202–215.
[18] A.-S. K. Pathan, Ed., in Security of self-organizing networks: MANET, WSN, WMN, VANET, Boca
Raton, Fla.: CRC Press, Auerbach, 2011, p. 200.
[19] S. Biswas, J. Misic, and V. Misic, “DDoS attack on WAVE-enabled VANET through
synchronization,” in Global Communications Conference (GLOBECOM), 2012, pp. 1079–1084.
[20] V. Bibhu, R. Kumar, B. S. Kumar, and D. K. Singh, “Performance Analysis of black hole attack in
VANET,” International Journal Of Computer Network and Information Security, vol. 4, no. 11, p. 47,
2012.
[21] Y.-C. Hu, A. Perrig, and D. B. Johnson, “Packet leashes: a defense against wormhole attacks in
wireless networks,” in INFOCOM 2003. Twenty-Second Annual Joint Conference of the IEEE
Computer and Communications. IEEE Societies, 2003, vol. 3, pp. 1976–1986.
[22] T. Moore, M. Raya, J. Clulow, P. Papadimitratos, R. Anderson, and J.-P. Hubaux, “Fast exclusion of
errant devices from vehicular networks,” in 2008 5th Annual IEEE Communications Society
Conference on Sensor, Mesh and Ad Hoc Communications and Networks, 2008, pp. 135–143.
[23] T. Leinmüller and E. Schoch, “Greedy routing in highway scenarios: The impact of position faking
nodes,” in Proceedings of Workshop On Intelligent Transportation (WIT 2006)(Mar. 2006), 2006.
[24] N.-W. Lo and H.-C. Tsai, “Illusion attack on VANET applications-A message plausibility problem,”
in Globecom Workshops, 2007 IEEE, 2007, pp. 1–8.
[25] M. Raya and J.-P. Hubaux, “Securing vehicular ad hoc networks,” Journal of Computer Security, vol.
15, no. 1, pp. 39–68, 2007.
[26] M. Jakobsson, S. Wetzel, and B. Yener, “Stealth attacks on ad-hoc wireless networks,” in Vehicular
Technology Conference, 2003. VTC 2003-Fall. 2003 IEEE 58th, 2003, vol. 3, pp. 2103–2111.
[27] P. Ning and K. Sun, “How to misuse AODV: a case study of insider attacks against mobile ad-hoc
routing protocols,” Ad Hoc Networks, vol. 3, no. 6, pp. 795–819, 2005.
[28] J. R. Douceur, “The sybil attack,” in International Workshop on Peer-to-Peer Systems, 2002, pp. 251–
260.
[29] P. Golle, D. Greene, and J. Staddon, “Detecting and correcting malicious data in VANETs,” in
Proceedings of the 1st ACM international workshop on Vehicular ad hoc networks, 2004, p. 29.
[30] B. Xiao, B. Yu, and C. Gao, “Detection and localization of sybil nodes in VANETs,” in Proceedings
of the 2006 workshop on Dependability issues in wireless ad hoc networks and sensor networks, 2006,
p. 1.
[31] B. Yu, C.-Z. Xu, and B. Xiao, “Detecting Sybil attacks in VANETs,” Journal of Parallel and
Distributed Computing, vol. 73, no. 6, pp. 746–756, Jun. 2013.
[32] G. Guette and B. Ducourthial, “On the Sybil attack detection in VANET,” in Mobile Adhoc and
Sensor Systems, 2007, pp. 1–6.

[33] T. Zhou, R. R. Choudhury, P. Ning, and K. Chakrabarty, “Privacy-preserving detection of sybil
attacks in vehicular ad hoc networks,” in Fourth Annual International Conference on Mobile and
Ubiquitous Systems: Networking & Services, 2007, pp. 1–8.
[34] S. Park, B. Aslam, D. Turgut, and C. C. Zou, “Defense against sybil attack in vehicular ad hoc
network based on roadside unit support,” in MILCOM 2009-2009 IEEE Military Communications
Conference, 2009, pp. 1–7.
[35] T. Zhou, R. R. Choudhury, P. Ning, and K. Chakrabarty, “P2DAP-Sybil Attacks Detection in
Vehicular Ad Hoc Networks,” IEEE Journal on Selected Areas in Communications, vol. 29, no. 3, pp.
582–594, Mar. 2011.
[36] Y. Hao, J. Tang, and Y. Cheng, “Cooperative sybil attack detection for position based applications in
privacy preserved VANETs,” in Global Telecommunications Conference (GLOBECOM 2011), 2011
IEEE, 2011, pp. 1–5.
[37] B. Lee, E. Jeong, and I. Jung, “A DTSA (Detection Technique against a Sybil Attack) Protocol Using
SKC (Session Key Based Certificate) on VANET,” Int. J. Security its Appl, vol. 7, no. 3, pp. 1–10,
2013.
[38] M. Rahbari and M. A. J. Jamali, “Efficient detection of sybil attack based on cryptography in
VANET,” International Journal of Network Security & Its Applications (IJNSA), vol. 3, no. 6, 2011.
[39] X. Feng, C. Li, D. Chen, and J. Tang, “A method for defensing against multi-source Sybil attacks in
VANET,” Peer-to-Peer Networking and Applications, vol. 10, no. 2, pp. 305–314, 2017.
[40] S. Chang, Y. Qi, H. Zhu, J. Zhao, and X. Shen, “Footprint: detecting Sybil attacks in urban vehicular
networks,” IEEE Transactions on Parallel and Distributed Systems, vol. 23, no. 6, pp. 1103–1114,
2012.
[41] C. Chen, X. Wang, W. Han, and B. Zang, “A robust detection of the sybil attack in urban vanets,” in
29th IEEE International Conference on Distributed Computing Systems Workshops, 2009, pp. 270–
276.
[42] J. Grover, M. S. Gaur, V. Laxmi, and N. K. Prajapati, “A sybil attack detection approach using
neighboring vehicles in VANET,” in Proceedings of the 4th international conference on Security of
information and networks, 2011, pp. 151–158.
[43] J. Soryal and T. Saadawi, “DoS attack detection in Internet-connected vehicles,” in 2013 International
Conference on Connected Vehicles and Expo (ICCVE), 2013, pp. 7–13.
[44] K. Verma, H. Hasbullah, and A. Kumar, “Prevention of DoS attacks in VANET,” Wireless personal
communications, vol. 73, no. 1, pp. 95–126, 2013.
[45] K. Verma and H. Hasbullah, “Bloom-filter based IP-CHOCK detection scheme for denial of service
attacks in VANET,” Security and Communication Networks, vol. 8, no. 5, pp. 864–878, 2015.
[46] C. A. Kerrache, N. Lagraa, C. T. Calafate, and A. Lakas, “TFDD: A trust-based framework for
reliable data delivery and DoS defense in VANETs,” Vehicular Communications, Dec. 2016.
[47] J. Zhang, P. Porras, and J. Ullrich, “Highly Predictive Blacklisting,” in Proceedings of the 17th
Conference on Security Symposium, Berkeley, CA, USA, 2008, pp. 107–122.
[48] J. Hortelano, J. C. Ruiz, and P. Manzoni, “Evaluating the usefulness of watchdogs for intrusion
detection in vanets,” in 2010 IEEE International Conference on Communications Workshops, 2010,
pp. 1–5.
[49] A. Daeinabi and A. G. Rahbar, “Detection of malicious vehicles (DMV) through monitoring in
Vehicular Ad-Hoc Networks,” Multimedia tools and applications, vol. 66, no. 2, pp. 325–338, 2013.
[50] M. Kadam and S. Limkar, “Performance Investigation of DMV (Detecting Malicious Vehicle) and
D&PMV (Detection and Prevention of Misbehave/Malicious Vehicles): Future Road Map,” in
Proceedings of the International Conference on Frontiers of Intelligent Computing: Theory and
Applications (FICTA), 2013, pp. 379–387.
[51] U. Khan, S. Agrawal, and S. Silakari, “Detection of malicious nodes (dmn) in vehicular ad-hoc
networks,” Procedia Computer Science, vol. 46, pp. 965–972, 2015.

[52] O. A. Wahab, H. Otrok, and A. Mourad, “A Dempster–Shafer Based Tit-for-Tat Strategy to Regulate
the Cooperation in VANET Using QoS-OLSR Protocol,” Wireless Personal Communications, vol. 75,
no. 3, pp. 1635–1667, 2014.
[53] O. A. Wahab, H. Otrok, and A. Mourad, “VANET QoS-OLSR: QoS-based clustering protocol for
Vehicular Ad hoc Networks,” Computer Communications, vol. 36, no. 13, pp. 1422–1435, Jul. 2013.
[54] R. Baiad, H. Otrok, S. Muhaidat, and J. Bentahar, “Cooperative cross layer detection for blackhole
attack in VANET-OLSR,” in 2014 International Wireless Communications and Mobile Computing
Conference (IWCMC), 2014, pp. 863–868.
[55] H. Sanadiki, H. Otrok, A. Mourad, and J.-M. Robert, “Detecting attacks in QoS-OLSR protocol,” in
2013 9th International Wireless Communications and Mobile Computing Conference (IWCMC),
2013, pp. 1126–1131.
[56] R. Baiad, O. Alhussein, H. Otrok, and S. Muhaidat, “Novel cross layer detection schemes to detect
blackhole attack against QoS-OLSR protocol in VANET,” Vehicular Communications, vol. 5, pp. 9–
17, Jul. 2016.
[57] Y. Guo, S. Schildt, J. Morgenroth, and L. C. Wolf, “A Misbehavior Detection System for Vehicular
Delay Tolerant Networks.,” in INFORMATIK, 2012, pp. 1871–1877.
[58] F. Li, J. Wu, and A. Srinivasan, “Thwarting blackhole attacks in disruption-tolerant networks using
encounter tickets,” in The 28th Conference on Computer Communications, 2009, pp. 2428–2436.
[59] Y. Guo, S. Schildt, and L. Wolf, “Detecting blackhole and greyhole attacks in vehicular delay tolerant
networks,” in 2013 Fifth International Conference on Communication Systems and Networks
(COMSNETS), 2013, pp. 1–7.
[60] Y. Guo, S. Schildt, and L. Wolf, “Using Cluster Analysis to Detect Attackers in Vehicular Delay
Tolerant Networks,” in International Conference on Ad Hoc Networks, 2013, pp. 181–196.
[61] P. Nagrath, S. Aneja, N. Gupta, and S. Madria, “Protocols for mitigating blackhole attacks in delay
tolerant networks,” Wireless Networks, vol. 22, no. 1, pp. 235–246, 2016.
[62] X. Yao, X. Zhang, H. Ning, and P. Li, “Using trust model to ensure reliable data acquisition in
VANETs,” Ad Hoc Networks, vol. 55, pp. 107–118, Feb. 2017.
[63] S. M. Safi, A. Movaghar, and M. Mohammadizadeh, “A novel approach for avoiding wormhole
attacks in VANET,” in 2009 First Asian Himalayas International Conference on Internet, 2009, pp.
1–6.
[64] S. Biswas and J. Misic, “A Cross-Layer Approach to Privacy-Preserving Authentication in WAVE-
Enabled VANETs,” IEEE Transactions on Vehicular Technology, vol. 62, no. 5, pp. 2182–2192, Jun.
2013.
[65] T. H.-J. Kim et al., “Vanet alert endorsement using multi-source filters,” in Proceedings of the seventh
ACM international workshop on VehiculAr InterNETworking, 2010, pp. 51–60.
[66] M. Raya, P. Papadimitratos, I. Aad, D. Jungels, and J.-P. Hubaux, “Eviction of misbehaving and
faulty nodes in vehicular networks,” IEEE Journal on Selected Areas in Communications, vol. 25, no.
8, pp. 1557–1568, 2007.
[67] M. Ghosh, A. Varghese, A. Gupta, A. A. Kherani, and S. N. Muthaiah, “Detecting misbehaviors in
VANET with integrated root-cause analysis,” Ad Hoc Networks, vol. 8, no. 7, pp. 778–790, Sep. 2010.
[68] A. Vora and M. Nesterenko, “Secure location verification using radio broadcast,” IEEE Transactions
on Dependable and Secure Computing, vol. 3, no. 4, pp. 377–385, 2006.
[69] J.-P. Hubaux, S. Capkun, and J. Luo, “The security and privacy of smart vehicles,” IEEE Security &
Privacy Magazine, vol. 2, no. LCA-ARTICLE-2004-007, pp. 49–55, 2004.
[70] S. Brands and D. Chaum, “Distance-bounding protocols,” in Workshop on the Theory and Application
of Cryptographic Techniques, 1993, pp. 344–359.
[71] T. Leinmuller, E. Schoch, and F. Kargl, “Position verification approaches for vehicular ad hoc
networks,” IEEE Wireless Communications, vol. 13, no. 5, pp. 16–21, 2006.
[72] N. Bismeyer, C. Stresing, and K. M. Bayarou, “Intrusion detection in VANETs through verification of
vehicle movement data,” in Vehicular Networking Conference (VNC), 2010 IEEE, 2010, pp. 166–173.

[73] R. K. Schmidt, T. Leinmüller, E. Schoch, A. Held, and G. Schäfer, “Vehicle behavior analysis to
enhance security in vanets,” in Proceedings of the 4th IEEE Vehicle-to-Vehicle Communications
Workshop (V2VCOM2008), 2008.
[74] S. Ruj, M. A. Cavenaghi, Z. Huang, A. Nayak, and I. Stojmenovic, “On data-centric misbehavior
detection in VANETs,” in Vehicular technology conference (VTC Fall), 2011 IEEE, 2011, pp. 1–5.
[75] A. Studer, M. Luk, and A. Perrig, “Efficient mechanisms to provide convoy member and vehicle
sequence authentication in VANETs,” in Third International Conference on Security and Privacy in
Communications Networks and the Workshops, 2007, pp. 422–432.
[76] C. Adjih, D. Raffo, and P. Mühlethaler, “Attacks against OLSR: Distributed key management for
security,” in 2nd OLSR Interop/Workshop, Palaiseau, France, 2005.
[77] G. Yan, S. Olariu, and M. C. Weigle, “Providing VANET security through active position detection,”
Computer Communications, vol. 31, no. 12, pp. 2883–2897, 2008.
[78] G. Yan, B. B. Bista, D. B. Rawat, and E. F. Shaner, “General active position detectors protect
VANET security,” in 2011 International Conference on Broadband and Wireless Computing,
Communication and Applications (BWCCA), 2011, pp. 11–17.
[79] N. Kumar and N. Chilamkurti, “Collaborative trust aware intelligent intrusion detection in VANETs,”
Computers & Electrical Engineering, vol. 40, no. 6, pp. 1981–1996, 2014.
[80] H. Sedjelmaci and S. M. Senouci, “An accurate and efficient collaborative intrusion detection
framework to secure vehicular networks,” Computers & Electrical Engineering, vol. 43, pp. 33–47,
2015.
[81] O. A. Wahab, A. Mourad, H. Otrok, and J. Bentahar, “CEAP: SVM-based intelligent detection model
for clustered vehicular ad hoc networks,” Expert Systems with Applications, vol. 50, pp. 40–54, 2016.
[82] J. Grover, N. K. Prajapati, V. Laxmi, and M. S. Gaur, “Machine learning approach for multiple
misbehavior detection in VANET,” in International Conference on Advances in Computing and
Communications, 2011, pp. 644–653.
[83] J. Grover, V. Laxmi, and M. S. Gaur, “Misbehavior Detection Based on Ensemble Learning in
VANET,” in Advanced Computing, Networking and Security, vol. 7135, P. S. Thilagam, A. R. Pais,
K. Chandrasekaran, and N. Balakrishnan, Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2012,
pp. 602–611.
[84] T. Bouali, S.-M. Senouci, and H. Sedjelmaci, “A distributed detection and prevention scheme from
malicious nodes in vehicular networks,” International Journal of Communication Systems, vol. 29, no.
10, pp. 1683–1704, 2016.
[85] C. A. Kerrache, N. Lagraa, C. T. Calafate, J.-C. Cano, and P. Manzoni, “T-VNets: A novel trust
architecture for vehicular networks using the standardized messaging services of ETSI ITS,”
Computer Communications, vol. 93, pp. 68–83, Nov. 2016.
[86] E. ETSI, “302 637-2 V1.3.1-Intelligent Transport Systems (ITS); Vehicular Communications; Basic
Set of Applications; Part 2: Specification of Cooperative Awareness Basic Service,” ETSI, Sept, 2014.
[87] E. ETSI, “302 637-3 V1.2.2 (2014-11) Intelligent Transport Systems (ITS),” Vehicular
Communications.
[88] K. Zaidi, M. B. Milojevic, V. Rakocevic, A. Nallanathan, and M. Rajarajan, “Host-Based Intrusion
Detection for VANETs: A Statistical Approach to Rogue Node Detection,” IEEE Transactions on
Vehicular Technology, vol. 65, no. 8, pp. 6703–6714, Aug. 2016.
[89] “WHO | Number of road traffic deaths,” WHO. [Online]. Available:
http://www.who.int/gho/road_safety/mortality/traffic_deaths_number/en/. [Accessed: 18-Jan-2017].
[90] S. X. Wu and W. Banzhaf, “The use of computational intelligence in intrusion detection systems: A
review,” Applied Soft Computing, vol. 10, no. 1, pp. 1–35, 2010.
[91] S. Axelsson, “Intrusion detection systems: A survey and taxonomy,” Technical report, 2000.
[92] J. Zhang, “A Survey on Trust Management for VANETs,” in 2011 IEEE international conference on
Advanced information networking and applications (AINA), 2011, pp. 105–112.

[93] Y. Ji, P. Yue, and Z. Cui, “VANET 2.0: Integrating Visible Light with Radio Frequency
Communications for Safety Applications,” in Cloud Computing and Security, vol. 10040, X. Sun, A.
Liu, H.-C. Chao, and E. Bertino, Eds. Cham: Springer International Publishing, 2016, pp. 105–116.
[94] A.-M. Cailean, B. Cagneau, L. Chassagne, V. Popa, and M. Dimian, “A survey on the usage of DSRC
and VLC in communication-based vehicle safety applications,” in 2014 IEEE 21st Symposium on
Communications and Vehicular Technology in the Benelux (SCVT), 2014, pp. 69–74.
[95] S. Sen, “A survey of intrusion detection systems using evolutionary computation,” in Bio-Inspired
Computation in Telecommunications, 2015, Chapter 4, pp. 73–94.
[96] S. Pastrana, A. Mitrokotsa, A. Orfila, and P. Peris-Lopez, “Evaluation of classification algorithms for
intrusion detection in MANETs,” Knowledge-Based Systems, vol. 36, pp. 217–225, 2012.