Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Received: 30 July 2019 Revised: 8 February 2020 Accepted: 9 March 2020

DOI: 10.1002/dac.4402

RESEARCH ARTICLE

Survey on vehicular ad hoc networks clustering algorithms:


Overview, taxonomy, challenges, and open research issues

Oussama Senouci1,2 Saad Harous3 Zibouda Aliouat1

1
LRSD laboratory, Department of
Computer Science, Ferhat Abbas – Sétif 1 Summary
University, Sétif, Algeria Vehicular ad hoc networks (VANETs) have recently attracted considerable atten-
2
Department of Computer Science,
tion owing to their wide range of applications. However, there are several
Mohamed El-Bachir El-Ibrahimi
University, Bordj Bou Arreridj, Algeria challenges, such as mobility, routing, scalability, quality of services, and security.
3
College of Information Technology, Clustering is an important control mechanism in high-mobility networks and
United Arab Emirates University, Al-Ain, has been verified to be a promising approach in VANETs as well, as it ensures a
United Arab Emirates
basic level of network performance. Accordingly, several clustering algorithms
Correspondence have been proposed for these networks, and different protocols typically focus on
Saad Harous, College of Information
various performance metrics. In this study, we provide a thorough review of clus-
Technology, United Arab Emirates
University, Al-Ain, UAE. tering algorithms in VANETs. First, we present background material regarding
Email: harous@uaeu.ac.ae the clustering process. Secondly, we propose a new taxonomy that categorizes
clustering algorithms in VANETs based on different design aspects and pro-
vides a description of the algorithms in each category. Thirdly, an analysis of the
algorithms in each category is provided according to various comparison met-
rics. Fourthly, we highlight the main challenges for each category and discuss
some open research issues. Finally, we provide a general comparison of differ-
ent clustering algorithms according to selected key parameters. Thus, this study
provides a more thorough understanding of VANET clustering algorithms and
the research trends in this area.

K E Y WO R D S
clustering, MANET, taxonomy, VANET

1 I N T RO DU CT ION

Vehicular ad hoc networks (VANETs) are a special subclass of ad hoc networks whose purpose is to enhance road safety by
providing information on traffic, accidents, dangers, possible deviations, or weather information. Considering the fluidity
of road traffic, these networks have attracted notable attention in both academia and industry owing to their potential
to improve road safety and the convenience they offer to drivers and passengers. Most studies on VANETs have focused
on specific problems and applications, such as data dissemination and aggregation, channel access management, traffic
safety, topology discovery, and, particularly, routing schemes.1
The design of an effective application is a major challenge that should not be neglected in VANETs, considering their
special features and characteristics, such as high vehicle mobility and quick topology changes, which make the design
and implementation of effective solutions for such networks a difficult task.2 First, high mobility is the main factor distin-
guishing VANETs from other classes of ad hoc and wireless networks. Vehicle speed varies according to road conditions
and may be low or medium in urban areas and large on highways. This speed variation has a direct impact on network
stability and results in a dynamic network topology. Secondly, node density in a VANET area is not uniform but exhibits

Int J Commun Syst. 2020;e4402. wileyonlinelibrary.com/journal/dac © 2020 John Wiley & Sons, Ltd. 1 of 21
https://doi.org/10.1002/dac.4402
2 of 21 SENOUCI ET AL.

spatiotemporal variation. Typically, the density in urban areas is higher than in rural areas and depends on the time of
day.3 Finally, network fragmentation generally occurs when vehicle density is low and irregular. Then, the vehicles move
in disconnected isolated clusters, and therefore, end-to-end communication becomes difficult.4
Considering all these issues, new techniques and mechanisms have been studied to ensure a stable network topol-
ogy structure as well as effective data routing and dissemination. Clustering is an important technique used in VANETs
and presents an interesting solution for simplifying and optimizing network functions and services. It has significantly
improved the performance in numerous applications compared with the conventional flat structure.5 The clustering pro-
cess aims to structure network nodes into small groups called clusters. Typically, geographically neighboring vehicles are
grouped in the same cluster according to various parameters and metrics. The present study provides a thorough review of
clustering algorithms in VANETs. First, we present a general overview of the clustering technique in VANETs. Specifically,
we provide background knowledge regarding the clustering process: brief historical notes, definitions, cluster structures,
cluster head election criteria, cluster formation strategy, general procedural flow, and performance evaluation metrics.
Secondly, we propose a new taxonomy for clustering algorithms in VANETs, with a detailed description of each algorithm.
Thirdly, a detailed comparison is provided for each category of the proposed taxonomy considering relevant key parame-
ters, such as vehicle density, cluster stability, latency, cluster overlapping, and clustering overhead. Fourthly, we highlight
the main challenges for each category and discuss some open research issues. Finally, we provide a general comparison
of different clustering algorithms according to selected parameters. The main contributions of this survey are as follows:
• We provide a general overview of clustering in VANETs, including definitions, cluster structure, CH selection criteria,
general procedural flow, and performance evaluation metrics.
• We provide a new taxonomy for classifying recently proposed clustering algorithms for VANETs.
• We provide a detailed description of each existing solution as well as a comparison for each class of the proposed
taxonomy considering relevant key parameters.
• We highlight the main challenges encountered for each category and discuss some open research issues.
• We provide a general comparison of different clustering algorithms according to a number of selected parameters.
The rest of this survey paper is organized as follows. Section 2 presents the survey methodology. Section 3 provides an
overview of the clustering process in VANETs. Section 4 presents related work. Section 5 discusses the proposed taxonomy
of clustering algorithms for VANETs in detail. Finally, the paper is concluded in Section 6.

2 S U RV E Y METH O D O LO GY

As several clustering algorithms have been proposed for VANETs, we will thoroughly survey some of them. The selected
algorithms were identified through a combination of existing literature reviews on clustering in VANETs6-10 with our
own search for clustering techniques that explicitly target VANET networks. Moreover, we considered recently proposed
clustering algorithms that were not included in the previous surveys. Subsequently, we reviewed the abstracts of the
corresponding papers to identify the outline of each algorithm. Our methodology was to use the search terms “VANET”
or “Vehicular ad hoc networks” and “clustering”. Furthermore, we limited the search only to academic articles written in
English.

3 CLUSTERING IN VA NETS: A N OVERVIEW

In recent years, clustering has become one of the most widely used control mechanisms in VANETs, so that their vari-
ous challenges may be overcome and their performance be improved. Bali et al.6 define clustering as the mechanism of
grouping vehicles into small groups called clusters based on predefined metrics, such as node density, average velocity,
and node location. This grouping is performed according to the requirements of the target application to provide an easily
manageable network.

3.1 Brief history


The DARPA packet-radio network11 is the earliest notable clustering application. The main goal was to form autonomous
subnets in a mobile ad hoc network (MANET) for data and resource sharing. Lin et al.12 introduced the widely used low-
est ID and highest degree (LID-HD) clustering protocols for MANETs. Mobility clustering9 and the weighted clustering
SENOUCI ET AL. 3 of 21

FIGURE 1 Example of clusters structure

algorithm (WCA)13 were later proposed in an attempt to introduce mobility and weighted metrics to the clustering pro-
cess. The significant improvements of these algorithms led to the design of several other algorithms, such as distributed
group mobility adaptive clustering14 and the mobility-aware highest degree (MobHiD).15 As VANETs are a subclass of
MANETs, the clustering algorithms designed for MANETs are widely used in VANETs. This was the beginning of using
clustering in vehicular networks.

3.2 Clustering concepts


3.2.1 Cluster structure
Clustering is a process that aims to group the nodes of a network into small groups, called clusters, thus providing the
network with a hierarchical structure. This structure satisfies certain requirements, such as scalability, load balancing,
network stability, and quality of service (QoS). Typically, geographically neighboring nodes have a high probability to
join the same cluster according to certain rules and metrics. Typically, the cluster structure has three main types of nodes
(vehicles): a Cluster Head (CH), Cluster Member (CM), and Cluster Gateway (CG),16 as shown in Figure 1. Moreover,
researchers used other special types of nodes, such as a Temporary CH (TCH), Temporary CM (TCM), and Secondary CH
(SCH).
• CH is responsible for cluster coordination and communication with other clusters and network infrastructures.
Moreover, the CH has other tasks, such as relaying information between nodes in the same cluster (intra-cluster com-
munication) or between different clusters (inter-cluster communication). Compared with other nodes in the cluster,
the CH has some additional functions, such as data aggregation and channel access management.
• CMs are ordinary nodes that join a cluster according to their characteristics and similarities. They are responsible for
sending their application-based information and data to the CH in specific time intervals. CMs in a cluster cannot
communicate directly with CMs or CHs from other clusters.
• CG is a node that provides inter-cluster communication and can communicate with neighboring clusters and transfer
data. Typically, its position is at the border of clusters.
• TCH is a special type of CH with a short lifespan. It is used in certain protocols in the literature17-19 to ensure an
effective choice of primary CH.
• TCM is a special type of CM with a short lifespan. It is utilized in some algorithms20,21 to increase the stability of the
clusters and decrease the rate of overlap between clusters.
• SCH is a passive CH, which will replace the active CH, in case it becomes unavailable or leaves the cluster. It has been
used in previous researches22-24 to improve clusters stability.

3.2.2 Cluster heads election criteria


Numerous metrics and criteria have been used for selecting CHs.25 We will classify clustering algorithms into six
categories as shown in Figure 2, which are differentiated by the type of these metrics:
4 of 21 SENOUCI ET AL.

FIGURE 2 CH election metrics

1. Neglected metrics: This category contains algorithms that select CHs without using any metrics. Typically, these
schemes use heuristic algorithms, which have one or both of the following objectives: finding a solution with
reasonable runtime or finding the optimal solution. Moreover, heuristic algorithms exhibit reasonable performances.26
2. Random metrics: This category contains early clustering algorithms for ad hoc and MANET networks. These
algorithms use metrics with random values, which are mostly not significant, such as node identifiers.27
3. Position metrics: This category includes all GPS-based algorithms using specific metrics related to node position,
such as geographical position, angle, and Euclidean distance.
4. Mobility metrics: This category includes all the algorithms that are based on information related to node mobility,
such as velocity, acceleration, sent/received signals, and power or link stability.28
5. Combined metrics: This category refers to weighted algorithms that combine multiple metrics of different types,
such as degree, k-degree, and density to select the set of CHs.
6. Destination metrics: This class includes all the clustering algorithms that are based on mobility information related
to the destination object, such as relative speed, direction, and current location.

3.2.3 Cluster formation strategy


According to the available literature, there are two main cluster formation strategies: distributed and centralized
strategies.
1. Distributed strategy: Many clustering protocols have used this strategy14,22,23,29 to form the clusters. After receiving
the necessary information from the neighborhood, each node in the network seeks its affiliation locally. Then, every
node in the network must confirm its affiliation by sending an affiliation message to the CH of the appropriate cluster.
2. Centralized strategy: In this type of strategy, a central node takes charge to form the clusters and affiliate the members
to the appropriate clusters. Many works24,30-32 are based on this strategy. They use the roadside unit (RSU) as a central
node for clusters' formation and members' affiliation. First, the RSUs diffuse beacon messages to inform the nodes in
the network. Then, each node sends an affiliation request to the appropriate RSU. Once the RSU receives the affiliations
requests, it performs the cluster formation phase in a centralized way. Finally, the RSU informs the nodes of their
affiliations and roles.

3.2.4 Cluster radius


The cluster radius represents the maximum distance that separates the CH and its members. Typically, it is expressed in
number of hops. Accordingly, we distinguish two main categories of algorithms: 1-hop and k-hop algorithms.33

3.3 Clustering procedural flow


Typically, the clustering process consists of two main phases: cluster establishment and cluster maintenance. A series of
basic procedures is involved in these two phases, which may need to be repeated according to the conceptual rules of the
SENOUCI ET AL. 5 of 21

algorithm so that network requirements may be met. According to Cooper et al.,8 the clustering process is composed of
five main steps: Neighborhood discovery, CH election, announcement, affiliation, and maintenance, as shown in Figure 3.

3.3.1 Neighborhood discovery


Initially, when a vehicle decides to join the network, it should turn on its communication system and update its state to an
unclustered node. Then, it should announce its existence. In this respect, the vehicle broadcasts a periodic hello message
to its neighborhood, while simultaneously receiving similar messages from the neighborhood. These messages contain
necessary information to perform the clustering process.

3.3.2 Cluster heads election


After receiving the packet information from the neighborhood, the vehicle will check this information to elect the suitable
node to act as its new CH. The main way for choosing the new CH depends on the metrics used by the clustering protocol.
If the node itself is better suited for being CH, it will update its state to CH and move to the announcement step; otherwise,
if the chosen CH is found within its neighbor list, the node will move to the affiliation step.

3.3.3 Announcement
Each new CH must announce its election. Therefore, it broadcasts an announcement packet to begin the cluster for-
mation process by receiving the affiliation requests from unclustered nodes. At the end of this step, it will move to the
maintenance step.

3.3.4 Affiliation
When a node receives the announcement packet from a CH, it compares this CH with the CH already elected in the
previous step. If they are the same, the node sends a reply message, updates its state to CM, and joins the cluster; otherwise,
the node ignores the event and waits for another announcement message or moves directly to the maintenance step to
join the appropriate cluster.

3.3.5 Maintenance
This step depends on whether the node has become a CH or CM:
1. CH: Each CH monitors its CMs through the exchange of periodic messages to record the presence of members in
the cluster. When a CM node moves out of the range of the cluster, the CH detects this event and removes immedi-
ately this node from its member list. In addition, when two neighbor clusters have large overlap, the cluster merging

FIGURE 3 Clustering procedural flow8


6 of 21 SENOUCI ET AL.

process will commence. Typically, the merging procedure produces two CHs for the final larger cluster obtained.
Therefore, the CH that has the largest number of cluster members is elected as new CH for the cluster obtained. On
the contrary, when the CH receives a join request from an unclustered node, it should examine if the node is suitable
for its cluster. If this is the case, the CH adds this node to its member list and informs the node by sending a reply
message.
2. CM: A CM node will periodically examine the communication link with its CH through periodic messages. If the link
is lost, the CM node should change its state to unclustered and try to join another cluster. On other hand, when an
unclustered node approaches a cluster, it sends a join request to the CH of this cluster. Then, if the node receives a
reply message from the CH, it joins the cluster; otherwise, the node should search for another cluster.

TABLE 1 Different orders of the clustering steps applied Algorithm [Steps order]
in the reviewed algorithms
CDS37 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
DWCM38 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
SCRP39 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
CoD40 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
QoS-VANET41 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
HCAR24 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
smartCloud42 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
LID12 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
HD43 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
LID-TDMA27 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
Adaptive-LID44 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
FL-ECC/D45 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
PCTIC46 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
PSCA47 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
CB-BDP48 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
MC-DRIVE49 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
CRBP50 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
FUZZY17 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
DMC51 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
AMACAD52 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
ALM53 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
MCA-V2I23 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
HTMS54 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
EHCP55 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
WCS56 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
ACA57 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
AWCP58 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
MFCA-IoV59 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
CADVO29 [Step 1], [Step 4],[ Step 2], [Step 3], [Step 5.1 || Step 5.2]
WECA-MR22 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
EWCA60 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
LICA61 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
DBR62 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
DACR63 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
EUCLID64 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
DSARV65 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
MOSIC66 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
SOCA68 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
RMAC69 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
ICH70 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
N-hop71 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
DMCNF33 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
DHCV72 [Step 1], [Step 2],[ Step 4], [Step 3], [Step 5.1 || Step 5.2]
VMaSC-LTE73 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
PMC74 [Step 1], [Step 2],[ Step 3], [Step 4], [Step 5.1 || Step 5.2]
SENOUCI ET AL. 7 of 21

Table 1 shows a general comparison of the clustering steps order of the reviewed algorithms.

3.4 Performance metrics for clustering algorithms


To validate a clustering algorithm, it should be evaluated according to various performance metrics. Generally, these
metrics are used in comparisons with other existing solutions. According to Fauziyyah et al.,34 Amudhavel et al.35 and
Guizani,36 we have the following groups of metrics:
• Messages complexity: It is the number of control messages received by the nodes so that a valid cluster
re-organization may be performed after a change in the network topology.
• Network overhead: It is the number of messages related to the clustering process and the resources used by every
node during network operation, such as bandwidth, energy, memory, and time.
• Cluster structure stability: Certain metrics in this group are described below.
• CH lifetime: It is the length of any maximal time interval during which a node is CH. The average CH
lifetime is calculated by dividing the total CH lifetime by the total number of state changes from CH to
another state.
• CM lifetime: It is the length of any maximal time interval during which a node is CM. The average CM
lifetime is calculated by dividing the total CM lifetime by the total number of state changes from CM to
another state.
• CH change number: It represents the rate of state changes from CH to another state.
• Cluster number: It is the number of clusters constructed during network operation. Small values improve the
efficiency of the clustering algorithm.
• Average size of cluster: It is the average number of CM nodes managed by a CH during the total network operation
time.
• Time complexity: It represents the time required to perform a valid clusterre-organization after a change in network
topology.
• Message delivery latency: Refers to the average delay or time required for a message to be transmitted from a source
to a destination.23

4 RELATED WORK

Several taxonomies for clustering algorithms have been proposed,6-10 where the algorithms are categorized based on
different metric parameters. According to various key parameters, Bali et al.6 proposed a classification into six cate-
gories: predictive clustering including three subcategories: position-based, destination-based, and lane-based algorithms;
backbone-based clustering including k-hop based algorithms, medium access control (MAC)-based clustering including
Institute of Electrical and Electronics Engineers (IEEE) 802.11 MAC-based, time-division multiple access (TDMA)-based,
and space-division multiple access (SDMA)-based algorithms; traditional clustering including active and passive algo-
rithms, hybrid clustering including intelligence-based, distributed, and driver behavior-based algorithms; and secure
clustering including authentication-based algorithms. The authors also provided a detailed description of each algorithm
as well as a comparison among categories with respect to various key parameters. According to the clustering application,
Cooper et al.8 proposed a classification into eight classes: general purpose, routing, channel access management, security,
QoS, traffic safety, and topology discovery applications, as well as a combination with cellular infrastructure applications.
A survey of clustering protocols in VANETs can be found in Yang et al.,7 where objectives, challenges, and issues were
discussed. Moreover, the authors compared these protocols according to various parameters, such as relative velocity,
node density, cluster size, hop distance, and cluster establishment methodology. Hosmani et al.9 surveyed certain clus-
tering algorithms for VANETs and classified them based on cluster formation and CH election parameters. Moreover, the
authors provided a comparison for these clustering algorithms based on various key metrics, such as node density, scal-
ability, velocity, cluster lifetime, and feasibility. Pal et al.10 introduced an analytical model to evaluate the performance
of cluster-based algorithms in VANETs according to three important parameters: throughput, packet delivery ratio, and
end-to-end delay. Finally, we summarize the previous survey papers related to ours in Table 2, which highlights the main
contributions.
8 of 21 SENOUCI ET AL.

TABLE 2 Summary of previous surveys on clustering algorithms in VANETs


Year Authors Literature Contributions
2014 Bali Clustering in vehicular ad ▸ Clustering in VANETs with respect to various challenges
et al. hoc networks: Taxonomy,
challenges and solutions6
▸ Detailed taxonomy for clustering algorithms in VANETs
▸ Comparison of different clustering approaches based on different key parameters.
2017 Cooper A Comparative Survey ▸ Discussion of basic concepts related to the clustering process
et al. of VANET Clustering
Techniques8
▸ A new taxonomy for clustering algorithms in VANETs is provided
▸ Comparison of various clustering algorithms based upon some predefined metrics
2015 Yang Clustering algorithm in ▸ According to various parameters, a number of clustering algorithm is surveyed
et al. VANETs: A survey7
▸ An analysis of different VANETs clustering algorithms is provided
2017 Hosmani Survey on cluster based ▸ Presentation of a taxonomy of typical clustering algorithms
et al. routing protocol in
VANET9
▸ Discussion of the challenging factors in VANET clustering
2018 Pal Analytical model for clus- ▸ An analytical model to evaluate the performance clustering algorithm in VANETs
et al. tered vehicular ad hoc net-
work analysis10
▸ Comparison of clustered VANET performance with that of the non-clustered scenario.
▸ Analytical results are validated with simulation results

FIGURE 4 Taxonomy of clustering


algorithms in VANETs

5 TA XO N O MY OF CLU ST E R IN G ALGORITHMS IN VEHIC ULAR AD HOC


NETWORKS

Based on the CH election criteria, neighborhood (cluster radius) and cluster formation strategy discussed above, we pro-
pose a new taxonomy for VANET clustering algorithms, as shown in Figure 4. Subsequently, we discuss in depth various
VANET clustering algorithms based on the proposed taxonomy. Figure 5 illustrates the percentage of the VANET clus-
tering algorithms under each class of the proposed taxonomy including CH election criteria, neighborhood, and cluster
formation strategy. Table 3 shows the various parameters used for comparing the clustering algorithms with their descrip-
tions. These parameters cover most of the key areas that need to be considered for designing clustering algorithms and
characterizing their efficiency. The parameters selected for comparison of clustering algorithms are vehicle density, cluster
stability, latency, overhead, and cluster convergence.
SENOUCI ET AL. 9 of 21

FIGURE 5 Percentage of
VANET clustering algorithms
under each class of the taxonomy

TABLE 3 Comparison parameters and their descriptions


Parameter Description
Density The average number of vehicle per unit of area
Stability Cluster stability provided by the algorithm
Latency Average delay taken for a message to be transmitted from a source to a destination
Overhead Number of control messages received by the vehicles for performing the clustering process.
Convergence Elapsed time for the clusters formation phase

5.1 Heuristic clustering algorithms


Heuristic optimization is a method whereby a solution for a given problem is sought among possible solutions. Usually,
these algorithms can rapidly find a solution close to the optimal.75 Moreover, they are not based on particular metrics
(metrics neglected). Accordingly, several studies have been concerned with the design of heuristic clustering solutions to
find suboptimal solutions in reasonable time. However, most of the existing heuristic clustering solutions are sensitive
to the initialization and cannot ensure high performance levels. These algorithms are applied in ad hoc networks, and
particularly in VANETs, by solving certain graph theory problems, such as spanning tree construction, graph traversals,
maximal independent sets, and dominated sets. Certain algorithms in this category are described below.
Graph domination is one of the problems for which heuristic algorithms are effective. Cha et al.37 presented a new
clustering scheme based on a Connected Dominating Set (CDS) to prevent and resolve the broadcast storm problem in
VANETs. The CDS scheme considers the high mobility and connectivity of vehicles and matches the dynamic topology of
VANETs. Moreover, the CDS algorithm is a typical information dissemination technique and is suitable for various VANET
applications, such as emergency notifications and traffic safety. Moreover, a virtual core network can be constructed by
10 of 21 SENOUCI ET AL.

TABLE 4 Comparison of heuristic based Algorithm Density Stability Latency Overhead Convergence
clustering algorithms CDS37 Medium Medium High Medium High
DWCM38 High Medium Medium Low High
SCRP39 Medium High High Medium Medium
smartCloud42 High Medium High Medium High
CoD40 Medium High High Low High
QoS-VANET41 Medium Medium Medium Low High
HCAR24 Low Medium Medium Medium Medium

searching for a CDS in the network graph. Therefore, clustering using a CDS-based virtual core network can significantly
improve VANET performance.
Yan et al.38 introduced a new distributed weighted clustering algorithm based on mobility metrics (DWCM). By con-
sidering the dominating set problem in graph theory, a distributed algorithm for cluster formation and CH selection was
proposed, where vehicles in the k-hop dominating set are selected as new CHs. Moreover, topology changes caused by
the high vehicle mobility are handled in the cluster maintenance phase. The main goal of DWCM is to construct and
maintain stable k-hop clusters without incurring overly high network overhead.
In Togou et al.,39 the authors developed a new stable CDS-based routing protocol (SCRP). It is based on a distributed
geographic location source forwarding technique that considers the vehicular network topology to select data routing
paths with low end-to-end delay. To this end, the SCRP algorithm constructs stable backbones over road segments by
using two metrics: vehicle velocity and spatial distribution. These backbones are connected at intersections via special
gateway nodes that ensure an up-to-date vehicular network topology and monitor the expected delay for data routing
on the road segments. According to this information, SCRP assigns weights to the road segments. Therefore, the road
segments with the lowest weights are selected to form the routing paths. Thereby, SCRP can avoid the local maximum
problem and ensure load balancing on all possible routing paths.
In Chinnasamy et al.,42 the authors proposed a new Minimum connected dominating set-based RSU allocation for
smartCloud vehicles in VANET. Each smartCloud vehicle sends periodic beacon packet (Coordinates) to nearest RSU.
This latter periodically checks smartCloud vehicle whether it belongs to a CDS. SmartCloud Vehicles in the CDS utilize
a shorter waiting period and retransmit coordinates to the same RSU.
The spanning tree construction problem can also be tackled by heuristic clustering algorithms. Kponyo et al.40 pro-
posed a VANET Cluster-On-Demand (CoD) Minimum Spanning Tree (MST) approach in which clusters are constructed
by considering the intracluster QoS. The main goal of this algorithm is that the clusters are constructed only to relay infor-
mation on variable traffic density within the CH coverage area. Moreover, the proposed algorithm that implements the
MST ensures that the size of the clusters constructed does not affect the QoS in the network.
In Krishnakumar et al.,41 the authors proposed a new QoS-enabled data dissemination scheme that uses an improved
Kruskal's algorithm for efficient data dissemination and acceptable QoS in hierarchical VANETs. The proposed algorithm
extracts the MSTs based on Kruskal's algorithm on each road segment, where the nodes are clustered according to a fuzzy
cmeans clustering technique by considering the intracluster QoS. Thereby, every spanning tree will have a CH that is
responsible for gathering data from the leaf nodes and forwarding it to other coordinator nodes and vice versa.
In Senouci et al.,24 the authors introduced a new heuristic clustering algorithm based on RSU called HCAR for VANET.
The RSU is responsible for performing the cluster formation phase based on a simple heuristic algorithm, using graph
theory concepts, such as node degree and adjacency matrix. Moreover, HCAR recovers the problem of the unavailability
of CH by electing a secondary CH.

5.1.1 Challenges and open research issues in heuristic clustering algorithms


Heuristic algorithms are a key solution for improved clustering performance in VANETs. However, most heuristic clus-
tering schemes suffer from sensitivity to the initialization and do not ensure high performance and QoS levels. Moreover,
these algorithms suffer from data congestion and the local maximum problem. Furthermore, they do not consider met-
rics related to node mobility, which significantly reduces efficiency. Table 4 compares the heuristic clustering algorithms
considered in this study according to certain key parameters. It can be concluded that the heuristic method show an
acceptable stability rate and overhead. However, latency and cluster convergence should be further improved in order for
heuristic clustering algorithms to be effectively used for vehicular networks.
SENOUCI ET AL. 11 of 21

Algorithm Density Stability Latency Overhead Convergence TABLE 5 Comparison of MANETs clustering
LID12 Low Low High High High algorithms
HD43 Medium Low Medium High High
LID-TDMA27 Low Medium High Medium High
Adaptive-LID44 Medium Low High Medium High
FL-ECC/D45 Low Medium High High High

5.2 Mobile ad hoc networks clustering algorithms


As VANETs are a subclass of MANETs, conventional clustering algorithms designed for MANETs may be applied in
VANETs with some modifications. The vehicles in a VANET are characterized by their high mobility, which implies a
highly dynamic topology. Moreover, vehicle position in the same geographical proximity does not imply the same mobility
patterns. Thus, to apply MANET algorithms to VANETs, the specific characteristics of VANETs, such as mobility, direc-
tion, and location, should be considered.16 Typically, these algorithms use metrics with random values, which are mostly
not significant, such as node identifiers. Certain algorithms in this category are described below.
A well-known algorithm is the lowest-ID (LID),12 where every node is assigned a unique ID. To announce its existence,
each node in the network broadcast a “hello” message containing its ID within a period called “Hello period” (HP). Then,
each node constructs a table that contains its neighbors' ID information. Finally, the node with the lowest ID is selected
as a new CH; nodes that can hear two or more CHs at the same time become gateways, and the remaining nodes are CMs.
In Gerla et al.,43 the authors presented the highest degree (HD) algorithm. To perform the clustering process, each node
periodically broadcasts a beacon message including its degree value to its direct neighbors so that its degree value may be
compared with that of its neighbors. The node with the highest connectivity rate in the neighborhood becomes the CH,
and the remaining neighboring nodes become members.
In Nguyen et al.,27 the authors attempted to apply the LID algorithm in a cluster-based TDMA system for VANETs. A
new mechanism for selecting CHs that inherited the time division in TDMA- and MAC-frame format was proposed in
Sheu and Lin.76 The algorithm uses the conventional LID algorithm to improve network latency and to reduce signif-
icantly the number of reclustering operations. Therefore, clusters are efficiently formed. Moreover, the algorithm uses
cooperative MAC for intercluster communication to avoid cluster collision and overlap.
Gavalas et al.44 introduced a new efficient distributed clustering scheme that uses certain mobility metrics to construct
stable clusters. CHs are initially selected according to the cost-efficient lowest-ID technique. In the clustering maintenance
stage, node IDs are re-assigned based on different mobility parameters. As a result, nodes with low mobility are assigned
low IDs and therefore are selected as CHs. Moreover, the proposed solution relatively improves total network overhead
because the broadcast duration is adjusted according to the network mobility model.
In Prasad et al.,45 the authors presented a new fuzzy based incorporated bunching algorithm (FL-ECC/D) for vitality
productive steering conventions in MANET. The proposed bunching scheme utilizes fuzzy rationale to elect CHs and
introduces a partition separate between them for a balanced CHs dispersion based on a secured zone. Partition separate
is determined adaptively depending on the number of nodes in the network.

5.2.1 Challenges and open research issues in mobile ad hoc networks clustering
algorithms
Although VANETs are a subclass of MANETs and several VANET clustering algorithms are derived from MANETs,22
there are several differences, such as high mobility, highly dynamic topology, and limited driving directions. Moreover, we
notice that MANET algorithms usually use arbitrary metrics and do not consider the mobility patterns of VANET nodes.
As a result, the application of MANET clustering algorithms in VANETs remains limited and inefficient, and it does not
meet the main requirements of these networks. Table 5 compares the MANET clustering algorithms considered in this
study. It can be concluded that cluster stability, latency, clustering overhead, and cluster convergence need to be further
investigated in order for MANET clustering algorithms to be effectively applied in vehicular environment.

5.3 Position-based clustering algorithms


Position-based clustering algorithms use location and topology metrics, such as geographic position and spatial variation,
to perform the clustering process. They are among the most effective solutions for clustering in VANETs. Accordingly,
several clustering position algorithms have been proposed and have greatly improved the performance of VANETs owing
to their ability to handle vehicle position variation.6 Certain position-based algorithms are described below.
12 of 21 SENOUCI ET AL.

TABLE 6 Comparison of position based clustering Algorithm Density Stability Latency Overhead Convergence
algorithms PCTIC46 Low High Low Medium Low
PSCA47 High High Medium High Medium
CB-BDP48 Low Medium Low High Low
MC-DRIVE49 Low Medium Low High Low
CRBP50 Medium High Low Medium Medium

Wang et al.46 proposed a new position-based clustering technique for ad hoc intervehicle communication (PCTIC) in
large multi-hop VANETs. The clusters are formed according to vehicle position, and priorities are assigned to vehicles and
traffic information. The algorithm s elects one node as CH for each cluster based on the minimal dominating set problem.
Cluster size is defined according to the maximum distance between the CHs and their members. Thus, the clusters are
independently monitored and dynamically reconfigured as vehicles move. As a result, vehicles can move during cluster
establishment and maintenance.
Liu et al.47 introduced a position-sensitive clustering algorithm (PSCA) that is based on road configuration and section
position information and dynamic timestamps. In PSCA, the CH periodically broadcasts a control message at a certain
time-stamp by direction-based broadcasting. Each cluster member should send its position information in real time to
the CH before the next time-stamp, and the CH stores and reuses this information in the next iteration of the clustering
process. The initial choice of CHs is performed according to the principle of “aARandomly” Competition, First Declare
“Win”. Moreover, PSCA provides a special maintenance mechanism that facilitates cluster reconfiguration if the CH
leaves the cluster or new vehicles want to join the cluster.
In Alloulche et al.,48 the authors presented a new cluster-based beacon dissemination process (CB-BDP), where the
main goal is to provide vehicles with more accurate information on the position of neighboring vehicles. The CB-BDP
algorithm is designed to achieve two objectives. One is that the proximity map should be as wide and accurate as possible.
The other is that this map is shared with vehicles located nearby, thus allowing coordinated and synchronized reactions
of nearby vehicles to hazardous situations. In addition, the CB-BDP algorithm uses a new cluster-based gathering-based
tagging process involving an optimized topology for sharing the information collected by the proximity maps.
In Maslekar et al.,49 a new CH selection policy for direction-based clustering, called modified clustering based on direc-
tion in vehicular environment (MC-DRIVE), was introduced. This dissemination algorithm is based on another clustering
protocol called clustering based on direction in vehicular environment (C-DRIVE).77 MC-DRIVE constructs clusters com-
posed of nodes close to the intersections and according to their future directions (left, right, straight, and half-turn) after
the next intersection. Moreover, cluster size is limited to a certain distance from the intersection. At this distance, every
node broadcasts a hello message including its ID, driving lane, and future direction provided that it finds another vehicle
in front of it.
To improve the efficiency of V2 V communications, efficient Clustering V2 V Routing Based on Particle Swarm Opti-
mization in VANETs (CRBP) is proposed in Bao et al.50 First, vehicles with same moving directions are identified and
the CHs are elected. Second, for the required routing optimization, the route particle and its speed coding rules, iteration
rules, and fitness function are designed. Finally, mechanisms that can significantly improve the routing efficiency in the
cluster and among clusters are introduced.

5.3.1 Challenges and open research issues position based clustering algorithms
Position-based algorithms present an effective solution for improved VANET performance. However, in areas with high
mobility and dynamic topology, position metrics become irrelevant, and position variation increases enormously with
time. Moreover, position-based algorithms assume that every vehicle is equipped with a GPS system, which is not
practically true. Furthermore, positioning services accuracy can significantly vary. Table 6 compares the position-based
clustering algorithms considered in this study. It can be concluded that the position based algorithms can provide an
acceptable stability rate, low latency, and fast convergence. However, the global clustering overhead should be further
improved in order for position based algorithms to be effectively utilized for VANETs.

5.4 Mobility-based clustering algorithms


As VANETs are characterized by high mobility, dynamic topology, and limited driving directions, several mobility-based
clustering algorithms have recently been proposed. These algorithms use metrics related to node mobility (mobility met-
SENOUCI ET AL. 13 of 21

rics), such as relative velocity, acceleration, and direction, to perform the clustering process. Certainly these algorithms
are discussed below.
Hafeez et al.17 proposed a new a fuzzy-logic-based clustering algorithm for VANETs, where CHs are selected according
to the relative speed and average distance from neighbor nodes. Moreover, the proposed protocol enhances cluster stability
through the selection of a secondary CH (to be used as CH when the primary one becomes unavailable). Furthermore,
the maintenance step is adaptable to driver behavior based on a fuzzy logic inference system. Therefore, the algorithm is
suitable for environments with high mobility.
In Ren et al.,51 the authors introduced an efficient dynamic mobility-based clustering (DMC) algorithm for forming a
stable core network for future data aggregation and dissemination. The proposed algorithm performs the clustering pro-
cess according to vehicle mobility patterns, considering certain metrics, such as moving direction, relative speed, relative
average distance, and link stability. Moreover, the algorithm involves a new dynamic cluster formation phase, and a “tem-
porary CH” method is proposed to improve the cluster formation phase. Furthermore, the algorithm introduces a “safe
distance threshold” to monitor the cluster size.
Morales et al.52 presented an adaptable mobility-aware clustering algorithm based on destination positions (AMACAD)
for accurately following the network mobility patterns and improving cluster stability. AMACAD considers certain desti-
nation metrics, such as current position, relative velocity, and final destination, to perform the clustering process. Thus,
the algorithm provides a natural model of location reference for improved cluster stability and network performance. The
information is disseminated by the clusters, improving latency, reliability, and data delivery ratio.
Souza et al.53 presented a new aggregate local mobility (ALM) clustering algorithm that aimed at prolonging network
lifetime in VANETs. The ALM algorithm uses an ALM mechanism for controlling and monitoring existing CHs. To select
the CHs, each vehicle calculates its mobility variance over all neighbors. Lower variance implies less mobility and more
stability. Thus, a vehicle with less variance relative to its neighbors is suitable as a new CH.
In Senouci et al.,23 the authors proposed a new Multi-hop Clustering Approach over Vehicle-to-Internet, namely
MCA-V2I. This scheme is based on the reasonable assumption that a vehicle can connect to the Internet via a special
infrastructure called a Road Side Unit Gateway. Once connected to the Internet, each vehicle can obtain and share the
necessary information about its Multi-hop neighbors to perform the clustering process. The clustering is performed using
a Breadth-first search (BFS) algorithm for traversing a graph based on a Mobility Rate that is computed according to
various mobility metrics, such as relative velocity, nodes' degree and link stability.
In El Sayed et al.,54 the authors proposed a Hierarchical Trust Management System called HTMS that relies on the trust
criteria computation for each vehicle. Moreover, HTMS introduces a systematic scheme based on trust evaluation, trust
propagation, and trust aggregation to compute the “trustworthiness” of vehicles. The trust evaluation approach calculates
a local trust score based on a behavioral analysis. Then, the proposed trust propagation model broadcasts the computed
local trust scores to neighboring vehicles using a platoon-based dissemination technique. Based on the local trust scores
and propagated trust opinions, HTMS computes a Global Trust score (GTS) using the Dempster–Shafer method. At the
end, the proposed system invokes an action module to reward or punish vehicles based on the already calculated GTS
score.
Dutta et al.55 introduced an efficient hierarchical clustering protocol called EHCP for multihop communication for
effective resource utilization and improving cluster stability. The proposed EHCP assumes that the vehicles are linked
to the Internet using roadside unit gateway. When the vehicles are linked to the Internet, every vehicle aggregates
information about its neighboring nodes and performs the clustering process to select appropriate CHs.

5.4.1 Challenges and open research issues in mobility-based clustering algorithms


Table 7 compares mobility based clustering algorithms. It can be concluded that mobility-based algorithms are more suit-
able for VANETs. Moreover, an acceptable latency with an improvement on cluster convergence is observed owing to
the minimization of relative mobility as well as the distance of the CHs to their CMs. Furthermore, the reviewed algo-
rithms are a significant improvement over other types of algorithms because they consider mobility characteristics and
better meet the requirements of VANETs. As a result, mobility-based clustering algorithms are more suitable for VANETs,
particularly in high mobility areas. However, clustering overhead and cluster stability should be further improved.

5.5 Weight-based clustering algorithms


To select CHs, weight-based clustering algorithms use multiple metrics rather than a single metric. This is achieved
through a function representing the weight of a node. The general formula of the weight W(i) for a given node i according
14 of 21 SENOUCI ET AL.

TABLE 7 Comparison of mobility based clustering Algorithm Density Stability Latency Overhead Convergence
algorithms FUZZY17 Low Medium Medium Medium Low
DMC 51, 51 Medium Medium Low High Low
AMACAD52 Low Medium Medium Medium Low
ALM53 Medium Low Low Medium Low
MCA-V2I23 Low Medium Low Medium Low
HTMS54 Medium Medium Low Medium Medium
EHCP55 Low Medium Medium High Low

to the metrics A(i), B(i) and C(i) is as follows:

W(i) = 𝛼 × A(i) + 𝛽 × B(i) + … + 𝛾 × C(i) (1)

With 𝛼, 𝛽, and 𝛾 are coefficients, where 𝛼 + 𝛽 + 𝛾 = 1.


Algorithms in this category aim to elect a set of CHs that better meet the requirements of the network environment. Sev-
eral such algorithms have been proposed for VANETs and are described below. A new weighted clustering TDMA-MAC
scheme (WCS) for VANETs is proposed in Xie et al.,56 When selecting the set of CHs, the WCS algorithm uses indexes and
an entropy weight that combines certain parameters, such as radio signal transmitting power constraints, vehicle energy
consumption, and vehicle mobility. Based on the TDMA-MAC technique, a realistic clustering channel access mechanism
is introduced for reducing the chance of collision and ensuring efficient end-to-end communication in VANETs.
Hadded et al.57 introduced a new angle-based clustering algorithm (ACA) for VANETs; it uses vehicle angular position
and direction to select the most stable vehicles as CHs. In ACA, two or more vehicles can belong to the same cluster if
and only if the angle between their velocity vectors is acute.
A multiobjective genetic-algorithm-based adaptive weighted clustering protocol (AWCP) for VANETs was designed in.58
It elects CHs based on weights that combine certain key parameters, such as highway ID and vehicle direction, location,
velocity, and connectivity. The main goal of AWCP is to improve the stability of the network topology. AWCP uses a genetic
multi-objective method whose inputs are the metrics for the calculation of the node weights. This method aims to provide
a stable cluster structure, improve the data delivery rate, and reduce the total clustering overhead.
In Khan et al.,59 the authors proposed a new mechanism based on moth flame clustering optimization for the Internet of
vehicles (MFCA-IoV). The authors introduced a moth flame optimizer, which is a nature-inspired process. The MFCA-IoV
algorithm optimizes the generation of clusters to ensure robust transmission.
Aadil et al.29 proposed a new clustering algorithm for the Internet of vehicles based on dragonfly optimization, called
CAVDO. The authors use a swarm-based multiobjective method to perform the clustering process for improved topology
stability in a dynamic environment. In addition, the CAVDO algorithm uses a new mobility aware model, which makes
the transmission range more dynamic.
In Senouci et al.,22 the authors introduced a new Efficient Weight based Clustering Algorithm using MR (WECA-MR)
for IoV, which aims to improve the clusters' stability and reduce the total communication overhead. To perform the clus-
tering process, the proposed algorithm uses both classic weighted metrics, such as degree, average distance, and a new
metric, called MR. It is a metric that combines mobility metrics: relative velocity and relative acceleration. Moreover,
WECA-MR scheme shows that the unavailability of CH is minimized when the main CH and another Backup CH (BCH)
are elected.
Bello et al.60 proposed an enhanced weight-based clustering algorithm (EWCA) to improve VANET performances.
EWCA approach considers any vehicle moving on the same road segment with the same road ID and within the trans-
mission range of its neighbor to be suitable for the cluster formation phase. To elect the CHs, EWCA algorithm identifies
several metrics based on vehicle mobility information. Based on its relevance, any vehicle is associated with a prede-
fined weight value. A vehicle with the largest weight value is elected as the primary CH (PCH). Moreover, EWCA scheme
introduced a secondary CH (SeCH) to act as a backup to the PCH for improving cluster stability.

5.5.1 Challenges and open research issues in weight-based clustering algorithms


Weight-based algorithms present an effective solution for improved VANET performance. Typically, these algorithms use
multimetric mechanisms to perform the clustering process. However, the simultaneous use of multiple parameters ren-
ders parameter tuning a nontrivial problem in such conditions. Table 8 compares the weight-based clustering algorithms
considered in this study. It can be concluded that the weight-based algorithms provide good network performances with
SENOUCI ET AL. 15 of 21

Algorithm Density Stability Latency Overhead Convergence TABLE 8 Comparison of weigh based clustering
WCS56 Medium Medium Medium High Low algorithms
ACA57 Low Medium Low Medium Low
AWCP 58, 58 High High Medium Medium Medium
MFCA-IoV59 Medium High Low High Low
CAVDO29 High High Low Medium Medium
WECA-MR22 Low Medium Medium Medium Low
EWCA60 High High Low High Medium

a good cluster stability, low latency, and fast clustering convergence. On the other hand, the total clustering overhead
should be further improved in order for weight-based algorithms to be effectively used for vehicular environment.

5.6 Destination-based clustering algorithms


Destination-based clustering algorithms consider metrics related to the final destination of vehicles, such as current geo-
graphical position, direction, and relative velocity, to perform the clustering process. The destination is known in advance
using GPS devices and navigation systems. Certain algorithms in this category are described below.
Farhan et al.61 proposed robust localization using cluster analysis (LICA) to improve the accuracy of GPS devices. The
LICA algorithm utilizes a new modified tri-lateration mechanism for constructing a set of possible coordinates (x, y) based
on a cluster analysis mechanism, allowing accurate data according to the given weights and resulting in improved location
estimation for each vehicle. Moreover, unreliable data from malicious vehicles are removed during the clustering process.
As a result, vehicles can collect reliable real-time data and forward them to other vehicles, and thus vehicle destinations
can be safely reached.
Sethi et al.62 presented a Destination Based Routing (DBR) algorithm for context-based clusters in VANETs. The
DBR protocol provides two mechanisms for minimizing data traffic and end-to-end communication in VANETs. First, a
context-based clustering technique is designed that considers various metrics regarding vehicle destination: cluster forma-
tion location, direction, relative velocity, interest list, and final destination. Furthermore, a DBR protocol is proposed for
these context-based clusters to improve intercluster communication. By using this context-based clustering that includes
an interest list, the overall end-to-end communication for relaying information from source to destination is significantly
improved.
Aravindhan et al.63 designed a destination-aware context-based routing (DACR) scheme using a soft computing clus-
tering algorithm for VANETs. The authors also introduced two soft computing approaches. First, they designed a hybrid
clustering method that combined geographic and context clustering. The aim of this hybrid clustering was to reduce
clustering overhead and avoid network congestion. Secondly, a destination-aware routing protocol was introduced for
intercluster communication to improve the overall packet delivery ratio and the end-to-end delay. The DACR protocol
improves significantly intercluster routing by using the optimal next forwarding vehicle selection.
A new clustering routing algorithm for VANETs based on the Euclidean distance is presented in Tian et al.64 ; it uses
location and direction information to group the vehicles into self-organized clusters. Thus, if the source and destination
vehicles are located in different clusters, the direction information is used to restrict the data routing path in the same
direction. This not only assists in finding a shorter and more stable route but also improves the flooding in control mes-
sage rooting. To perform the clustering process, each vehicle announces its existence by broadcasting a beacon message
including its geographical position, direction, and current time. After receiving the beacon messages from its neighbors,
every vehicle in the network updates its topology table. Then, the CHs are generated by selecting the vehicles with the
minimum distance parameter.
Azat et al.65 introduced a new Destination based Stable clustering Algorithm and Routing for VANET (DSARV). The
proposed approach considers various physical parameters for cluster formation process such as vehicle location, direction,
velocity, and destination, as well as a possible list of interests of the vehicle. Based on these five parameters, DSARV
calculates the CH Eligibility of each Vehicle. Moreover, based on this CH Eligibility, it triggers the CH selection process.

5.6.1 Challenges and open research issues in destination-based clustering algorithms


Destination algorithms use various metrics related to destination to form stable clusters. Specifically, the ratio of cluster
head changes is significantly reduced, resulting in more stable clusters. This reduction is justified by the fact that a vehi-
cle can leave the cluster only when it encounters a CH whose destination is more similar to the destination of the current
16 of 21 SENOUCI ET AL.

TABLE 9 Comparison of destination based clustering Algorithm Density Stability Latency Overhead Convergence
algorithms LICA61 Medium Medium Medium High High
DBR62 Low Medium Medium Medium Medium
DACR63 Medium Medium Medium Medium Medium
EUCLID64 High High Low High High
DSARV65 Low Medium Medium High High

CH. Moreover, destination-based algorithms consider metrics related to vehicle destination thus significantly improving
network performance.6 These algorithms have certain drawbacks. One is the large transmission delay and global net-
work overhead when the density of vehicles becomes high. This adversely affects network performance. Moreover, these
algorithms assume that each vehicle is equipped with a GPS system, which is not practically true. Table 9 compares the
destination-based clustering algorithms considered in this study. It can be concluded that clustering overhead and clus-
ter convergence should be further improved in order for destination-based clustering algorithms to be effectively used for
vehicular networks.

5.7 Neighborhood-based clustering algorithms


According to the cluster radius, clustering algorithms can be divided into two main categories: 1-hop and k-hop clustering
algorithms. Some of these algorithms construct clusters with 1-hop distance, where each node is one hop from its CH,
whereas others generate clusters with k-hop distance, where every node is at a distance of at most k hops from its CH.

5.7.1 1-hop clustering algorithms


Ahwazi et al.66 presented a new mobility-aware single-hop clustering scheme (MOSIC) for VANETs suitable for highway
environments. The MOSIC algorithm presents a new clustering methodology that uses a Gauss–Markov mobility model
(GMM)67 for mobility prediction. This enables vehicles to predict their mobility relative to their neighbors. Moreover,
the GMM uses a memory-based mobility model that can calculate the next-hop vehicle position according to its current
mobility metrics.
In Caballero-gil et al.,68 the authors proposed a new self-organized clustering algorithm (SOCA) for VANETs; it is based
on 1-hop clustering to reduce network overhead in dense road traffic areas and maintain communication security by
combining public and secret keys. To select the CHs, SOCA uses an improved algorithm for the independent set problem
with a secret-key agreement technique based on the generalized Diffie–Hellman protocol. Moreover, SOCA constructs
a dynamic virtual backbone in the vehicular network. This backbone network is composed of all CHs and CGs already
selected in the cluster-formation phase.
Goonewardene et al.69 presented a new clustering algorithm called robust mobility adaptive clustering (RMAC) for
highly dynamic VANETs in future ITS. The RMAC algorithm uses a new vehicle priority method for adaptively and easily
identifying 1-hop neighbors and selecting optimal CHs according to certain metrics, such as vehicle position, direction,
and relative speed. Moreover, RMAC introduces a new concept called zone of interest that assists vehicles in maintaining
their neighbor table.
In Abbas et al.,70 the authors introduced a novel Intelligent Cluster-Head (ICH) to mitigate the handover problem of
clustering in VANETs. ICH approach introduces a new recover mechanism, which is a controller on two clusters that are
used to change uplink between clusters to solve the handover problem in the overlapping area.

5.7.2 Challenges and open research issues in 1-hop clustering algorithms


These algorithms provide a highly effective coordination to CHs and more reliable intracluster communication. However,
their cluster coverage area is small, leading to a large number of clusters and high maintenance overhead. Therefore, the
overlapping rate increases excessively. Table 10 compares the 1-hop based clustering algorithms considered in this study.
It can be concluded that 1-hop clustering algorithms provide an acceptable performances in term of cluster stability and
latency; however, clustering convergence needs further improving.

5.7.3 k-hop clustering algorithms


In Zhang et al.,71 the authors designed a new multihop clustering scheme for VANETs called N-hop. This is considered
the first algorithm based on multihop clustering for VANETs. The N-hop algorithm selects vehicles with low aggregate
SENOUCI ET AL. 17 of 21

Algorithm Density Stability Latency Overhead Convergence TABLE 10 Comparison of 1-hop based clustering
MOSIC66 Low Medium Medium Medium High algorithms
SOCA68 Medium Medium Low Low High
RMAC69 Low Medium Medium Medium High
ICH70 Medium Medium Low High High

Algorithm Density Stability Latency Overhead Convergence TABLE 11 Comparison of k-hop clustering
N-hop71 Low Medium Low High Medium algorithms
DMCNF33 High Medium Medium Medium Low
DHCV72 Medium Medium Medium High Low
VMaSC-LTE73 High High Low High Low
PMC74 Medium High Low Medium Low

mobility as CHs because including the low aggregate mobility metric is expected to facilitate the construction of more
stable clusters and topology.
Chen et al.33 proposed a distributed multihop clustering algorithm for VANETs based on neighborhood follow
(DMCNF); it selects CHs according to the neighborhood follow relationship among vehicles. In the DMCNF algorithm,
a vehicle cannot directly identify the most suitable CH among its multihop neighbors; however, it knows the most stable
and similar 1-hop neighbors, and therefore, they probably belong to the same cluster. Accordingly, a vehicle can select its
CH by following the most stable vehicle. This technique is called neighborhood follow relationship.
Azizian et al.72 introduced a new distributed D-hop clustering algorithm, called DHCV, that constructs stable clusters
based on relative mobility information. DHCV uses the velocity and the location differences of vehicles as metrics to
perform the clustering process in the D-hop communication range. To form multihop clusters, every vehicle in DHCV
selects its CH according to the relative mobility within its D-hop neighbors.
A new multihop cluster-based IEEE 802.11p and long-term evolution (LTE) hybrid architecture for VANET safety mes-
sage dissemination, namely, VMaSC-LTE, is proposed in Ucar et al.73 The authors claim that in networks based only
on LTE technology, the delivery ratio and end-to-end delay for safety message relaying are degraded owing to broadcast
storms and network disconnections. Moreover, the authors mention that basic cellular VANET communication is not fea-
sible owing to the high installation and maintenance cost. Moreover, the high vehicle mobility results in a large number
of handoff occurrences at the base station. To achieve a high data packet delivery ratio and improve latency, the authors
propose a new hybrid architecture combining multihop clustering based on IEEE 802.11p and LTE.
Zhang et al.74 introduced a novel Passive Multi-hop Clustering algorithm (PMC) to improve VANETs performances. The
PMC approach is based on the idea of a multihop clustering method that ensures cluster stability and coverage. During
CH election phase, a priority-based neighbor-following strategy is presented to elect the optimal neighbor vehicles to join
the same cluster. According to this strategy, the intercluster communications have high reliability and stability.

5.7.4 Challenges and open research issues in k-hop clustering algorithms


Several k-hop clustering algorithms for VANETs have been proposed in the last few years. Compared with 1-hop cluster-
ing algorithms, these algorithms can extend cluster coverage area, decrease the number of clusters, and improve cluster
stability. However, some issues remain in k-hop clustering for VANETs. For example, intracluster communication and
maintenance overhead should be further improved. Table 11 compares the k-hop clustering algorithms considered in
this survey. It can be concluded that k-hop clustering algorithms provide an acceptable performances in term of clus-
ter stability, latency, and clustering convergence; however, clustering overhead should be further improved in future
researches.

5.8 General comparison of clustering algorithms in vehicular ad hoc networks


This section provides a general comparison of different clustering algorithms in VANETs. In this respect, Table 12 presents
a comparative analysis of various clustering techniques based on a number of parameters used for clustering in VANETs.
These parameters cover most of the key areas that need to be considered for designing a clustering algorithm. The param-
eters selected for comparison of clustering are CH selection metrics, scalability, cluster radius, load balancing, complexity,
and QoS. According to Table 12, the heuristic algorithms provide a good load balancing, volatile scalability, and high
complexity. The MANET-based schemes do not support the scalability, with a bad load balancing and an acceptable
18 of 21 SENOUCI ET AL.

TABLE 12 General Algorithm CHs selection Scalability Radius Load balancing Complexity QoS
comparison of clustering
Heuristic clustering algorithms
algorithms in VANETs CDS37 Neglected Low 1-hop Moderate Moderate No
DWCM38 Neglected Moderate k-hops Good High No
SCRP39 Neglected Low 1-hop Good Moderate No
smartCloud42 Neglected High 1-hop Good High Yes
CoD40 Neglected Moderate 1-hop Good Moderate Yes
QoS-VANET41 Neglected Low 1-hops Good High Yes
HCAR24 Neglected High k-hops Good High Yes
MANETs clustering algorithms
LID12 Random Low 1-hop Bad Low No
HD43 Random Low 1-hop Bad Low No
LID-TDMA27 Random Moderate 1-hop Moderate Moderate No
Adaptive-LID44 Random Moderate 1-hop Bad Moderate No
FL-ECC/D45 Random Moderate 1-hop Bad Low No
Position-based clustering algorithms
PCTIC46 Position High k-hops Good High No
PSCA47 Position Moderate 1-hop Moderate Moderate No
CB-BDP48 Position High 1-hop Good Low No
MC-DRIVE49 Position Moderate k-hops Good Moderate Yes
CRBP50 Position Moderate 1-hop Moderate High Yes
Mobility-based clustering algorithms
FUZZY17 Mobility Low 1-hop Moderate High No
DMC51 Mobility Moderate 1-hop Bad Moderate No
AMACAD52 Mobility Low 1-hop Moderate Low No
ALM53 Mobility Moderate 1-hop Good Moderate Yes
MCA-V2I23 Mobility Moderate k-hop Good Moderate Yes
HTMS54 Mobility High 1-hop Good High Yes
EHCP55 Mobility Moderate k-hop Good Moderate Yes
Weight-based clustering algorithms
WCS56 Weight High 1-hop Good High Yes
ACA57 Weight Moderate k-hops Moderate Moderate No
AWCP58 Weight High k-hops Good High Yes
MFCA-IoV59 Weight Moderate 1-hop Good High Yes
CAVDO29 Weight High 1-hop Good High Yes
WECA-MR22 Weight High 1-hop Good Moderate No
EWCA60 Weight Moderate 1-hop Good High No
Destination-based clustering algorithms
LICA61 Destination Moderate k-hops Good High No
DBR62 Destination High k-hops Moderate High Yes
DACR63 Destination High k-hops Good Moderate Yes
EUCLID64 Destination Moderate 1-hop Bad Low No
DSARV65 Destination Moderate k-hops Bad Moderate No
1-hop clustering algorithms
MOSIC66 Neighbors Moderate 1-hop Moderate High No
SOCA68 Neighbors Low 1-hop Good High Yes
RMAC69 Neighbors Moderate 1-hop Moderate Moderate No
ICH70 Neighbors Low 1-hop Bad Moderate No
k-hop clustering algorithms
N-hop71 Neighbors Low k-hops Bad Low No
DMCNF33 Neighbors High k-hops Good Moderate Yes
DHCV72 Neighbors Moderate k-hops Moderate Moderate No
VMaSC-LTE73 Neighbors High k-hops Good High Yes
PMC74 Neighbors Moderate k-hops Good High Yes

complexity. The position-based algorithms support the scalability, provide a good load balancing and volatile complex-
ity. Mobility-based algorithms depict a bad scalability rate and volatile load balancing and complexity. Weight-based
approaches extensively support scalability with a good load balancing and high complexity. Destination-based schemes
SENOUCI ET AL. 19 of 21

provide an acceptable scalability rate and volatile load balancing and complexity. 1-hop-based algorithms have a bad
scalability, good load balancing, and high complexity. Finally, k-hop-based techniques show a volatile scalability, load
balancing, and complexity.

6 CO N C LU S I O N

This paper presented a thorough review of clustering algorithms in VANETs. First, we provided an overview of the clus-
tering technique in VANETs. We presented background knowledge regarding the clustering process, a brief historical
note, definitions, cluster structure, CH selection metrics, general procedural flow, and performance evaluation met-
rics. Secondly, we proposed a new taxonomy for clustering algorithms for VANETs with a detailed description of each
algorithm. Thirdly, a detailed comparison was provided for each category of the proposed taxonomy considering relevant
key parameters, such as vehicle density, cluster stability, latency, cluster overlapping, and clustering overhead. Fourthly,
we highlighted the main challenges for each category and discussed some open research issues. Finally, we provided a
general comparison of different clustering algorithms according to selected parameters. With this survey, readers can have
a more thorough understanding of VANET clustering algorithms and the research trends in this area.

ORCID
Oussama Senouci https://orcid.org/0000-0002-5345-0713
Saad Harous https://orcid.org/0000-0001-6524-7352

REFERENCES
1. Cheng J, Cheng J, Zhou M, Liu F, Gao S, Liu C. Routing in internet of vehicles: a review. IEEE T Intell Transp Sys. 2015;16(5):2339-2352.
2. Rayeni MS, Hafid A. Routing in heterogeneous vehicular networks using an adapted software defined networking approach. In: 2018 Fifth
International Conference on Software Defined Systems (SDS). Barcelona, Spain; 2018:25-31.
3. Cao Z, Shi K, Song Q, Wang J. Analysis of between vehicle density and network congestion in vanets. In: 2017 7th IEEE International
Conference on Electronics Information and Emergency Communication (ICEIEC). Macau, China; 2017:409-412.
4. Senouci O, Aliouat Z, Harous S. A review of routing protocols in internet of vehicles and their challenges. Sens Rev. 2019;39(1):58-70.
5. Al-Sultan S, Al-Doori MM, Al-Bayatti AH, Zedan H. A comprehensive survey on vehicular ad hoc network. J Net Comp Appl.
2014;37:380-392.
6. Bali RS, Kumar N, Rodrigues J. Clustering in vehicular ad hoc networks: Taxonomy, challenges and solutions. V Commu.
2014;1(3):134-152.
7. Yang P, Wang J, Zhang Y, Tang Z, Song S. Clustering algorithm in VANETs: A survey. In: 2015 IEEE 9th International Conference on
Anti-counterfeiting, Security, and Identification (ASID). Xiamen, China; 2015:166-170.
8. Cooper C, Franklin D, Ros M, Safaei F, Abolhasan M. A comparative survey of VANET clustering techniques. IEEE Commu Surv Tut.
2017;19(1):657-681.
9. Hosmani S, Mathpati B. Survey on cluster based routing protocol in VANET. In: 2017 International Conference on Electrical, Electronics,
Communication, Computer, and Optimization Techniques (ICEECCOT). Mysuru, India; 2017:1-6.
10. Pal R, Prakash A, Tripathi R, Singh D. Analytical model for clustered vehicular ad hoc network analysis. ICT Expr. 2018;4(3):160-164.
11. Ephremides A, Wieselthier JE, Baker D. J.. A design concept for reliable mobile radio networks with frequency hopping signaling. Proc
IEEE. 1987;75(1):56-73.
12. Lin CR, Gerla M. Adaptive clustering for mobile wireless networks. IEEE J Select Area Commun. 1997;15(7):1265-1275.
13. Chatterjee M, Das SK, Turgut D. Wca: A weighted clustering algorithm for mobile ad hoc networks. Clust Comput. 2002;5(2):193-204.
14. Zhang Y, Ng JM. A distributed group mobility adaptive clustering algorithm for mobile ad hoc networks. In: 2008 IEEE International
Conference on Communications. Beijing, China; 2008:3161-3165.
15. Konstantopoulos C, Gavalas D, Pantziou G. Clustering in mobile ad hoc networks through neighborhood stability-based mobility
prediction. Computer Networks. 2008;52(9):1797-1824.
16. Sood M, Kanwar S. Clustering in MANET and VANET: a survey. In: 2014 International Conference on Circuits, Systems, Communication
and Information Technology Applications (CSCITA). Mumbai, India; 2014:375-380.
17. Hafeez K. A., Zhao L, Liao Z, Ma B. N.. A fuzzy-logic-based cluster head selection algorithm in VANETs. In: 2012 IEEE International
Conference on Communications (ICC). Ottawa, ON, Canada; 2012:203-207.
18. Alsuhli G, Khattab A, Fahmy Y. Double-head clustering for resilient VANETs. W Commu M Comp. 2019;2019:1-17.
19. Sulistyo S, Alam S, Adrian RJ. Coalitional game theoretical approach for VANET clustering to improve SNR. J Comp Netw and Communic.
2019;2019:4573619:1-4573619:13.
20. Rawashdeh Y, Mahmud S. A novel algorithm to form stable clusters in vehicular ad hoc networks on highways. EURASIP J Wire Commu
Netw. 2012;2012(1):15.
20 of 21 SENOUCI ET AL.

21. Cheng X, Huang B. A center-based secure and stable clustering algorithm for vanets on highways. W Commu Mob Comp. 2019;2019:1-10.
22. Senouci O, Harous S, Aliouat Z. An efficient weight-based clustering algorithm using mobility report for IoV. In: 2018 9th IEEE Annual
Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON) (IEEE UEMCON 2018). New York, USA; 2018;
New York, USA.
23. Senouci O, Aliouat Z, Harous S. Mca-v2i: A multi-hop clustering approach over vehicle-to-internet communication for improving vanets
performances. F Gen Comput Syst. 2019;96:309-323.
24. Senouci O, Harous S, Aliouat Z. A new heuristic clustering algorithm based on RSU for internet of vehicles. Arab J Sci Eng.
2019;44(11):9735-9753. https://doi.org/10.1007/s13369-019-03854-2
25. Mehta S, Sharma P, Kotecha K. A survey on various cluster head election algorithms for MANET. In: 2011 Nirma University International
Conference on Engineering. Ahmedabad, Gujarat, India; 2011:1-6.
26. Maizate A, Kamoun NE. A new metric based cluster head selection technique for prolonged lifetime in wireless sensor networks. Intern
J Commu Ant Prop (IRECAP). 2013;3(4):227-236.
27. Nguyen V, Kim O, Dang D, Kim S, Hong C. Application of the lowest-id algorithm in cluster-based tdma system for vanets. In: 2015
International Conference on Information Networking (ICOIN). Cambodia, Cambodia; 2015:25-30.
28. Fan W, Shi Y, Chen S, Zou L. A mobility metrics based dynamic clustering algorithm for vanets. IET Intern Conf Commun Techno Appl
(ICCTA 2011). 2011;2011:752-756.
29. Aadil F, Ahsan W, Rehman Z, Shah PA, Rho S, Mehmood I. Clustering algorithm for Internet of vehicles (IOV) based on dragonfly
optimizer (cavdo). J Supercomp. 2018;74(9):4542-4567.
30. Tomar R, Verma S, Tomar G. Cluster based RSU centric channel access for vanets. In: Gavrilova ML, Tan CJK, eds. Transactions on
computational science xvii. Berlin, Heidelberg: Springer Berlin Heidelberg; 2013:150-171.
31. Calvo JAL, Mathar R. A two-level cooperative clustering scheme for vehicular communications. In: 2016 6th International Conference on
Information Communication and Management (ICICM). Hatfield, UK; 2016:205-210.
32. Wang C, Li X, Li F, Lu H. A mobility clustering-based roadside units deployment for VANET. In: The 16th Asia-Pacific Network Operations
and Management Symposium. Hsinchu, Taiwan; 2014:1-6.
33. Chen Y, Fang M, Shi S, Guo W, Zheng X. Distributed multi-hop clustering algorithm for VANETs based on neighborhood follow. EURASIP
J Wirel Commun Netw. 2015;2015(1):1-12.
34. Fauziyyah AK, Sulistyo S, Mustika IW. Performance analysis of cluster formation method in vehicular ad-hoc networks. In: 2017 7th
International Annual Engineering Seminar (InAES). Yogyakarta, Indonesia; 2017:1-6.
35. Amudhavel J, Kumar KP, Smrithi RS, Banumathi S, Rajaguru D, Vengattaraman T. Performance evaluation of dynamic clustering of
vehicles in VANET: Challenges and solutions. In: Proceedings of the 2015 International Conference on Advanced Research in Computer
Science Engineering & Technology (ICARCSET 2015). Unna, India; 2015.
36. Guizani B. Algorithme de clusterisation et protocoles de routage dans les reseaux ad hoc. Ph.D. Thesis; 2012.
37. Cha S, Ryu M, Kim K, Jeon B. Applying connected dominating set to broadcasting in vehicular ad hoc networks. In: 2013 International
Conference on Information Science and Applications (ICISA). Suwon, South Korea; 2013:1-2.
38. Shi Y, Xu X, Lu C, Chen S. Distributed and weighted clustering based on d-hop dominating set for vehicular networks. KSII Transac Intern
Inform Syst. 2016;10:1661-1678.
39. Togou MA, Hafid A, Khoukhi L. Scrp: Stable CDS-based routing protocol for urban vehicular ad hoc networks. IEEE Transactions on
Intelligent Transportation Systems. 2016;17(5):1298-1307.
40. Kponyo JJ, Kuang Y, Zhang E, Domenic K. VANET cluster-on-demand minimum spanning tree (MST) prim clustering algorithm. In: 2013
International Conference on Computational Problem-Solving (ICCP). Jiuzhai, China; 2013:101-104.
41. Krishnakumar KG, Thomson Fredrik EJ. QoS enabled data dissemination in hierarchical VANET using machine learning approach. Intern
J Sig Imag Syst Eng. 2017;10:231.
42. Chinnasamy A, Sivakumar B, Selvakumari P, Suresh A. Minimum connected dominating set based RSU allocation for smartcloud vehicles
in vanet. Clust Comput. 2019;22(5):12,795-12,804. https://doi.org/10.1007/s10586-018-1760-8
43. Gerla M, Tzu-Chieh Tsai J. Multicluster, mobile, multimedia radio network. W Netw. 1995;1(3):255-265.
44. Gavalas D, Pantziou G, Konstantopoulos C, Mamalis B. Lowest-id with adaptive id reassignment: a novel mobile ad-hoc networks
clustering algorithm. In: 2006 1st International Symposium on Wireless Pervasive Computing. Phuket, Thailand; 2006:5-5.
45. Prasad AY, Balakrishna R. Energy efficient and secured clustering algorithm using fuzzy logic with k-means method in manet. Indian J
Sci Technol. 2019;12(19):292-301.
46. Wang Z, Liu L, Zhou M, Ansari N. A position-based clustering technique for ad hoc intervehicle communication. IEEE Trans Sys, Man,
Cybern, Part C (Applications and Reviews). 2008;38(2):201-208.
47. Liu H, Yang L, Zhang Y, Wu L. A position sensitive clustering algorithm for VANET. Intern J Onli Eng (iJOE). 2014;10:14.
48. Allouche Y, Segal M. Cluster-based beaconing process for vanet. V Commu. 2015;2(2):80-94.
49. Maslekar N, Mouzna J, Labiod H, Devisetty M, Pai M. Modified C-DRIVE: clustering based on direction in vehicular environment. In:
2011 IEEE Intelligent Vehicles Symposium (IV). Baden-Baden, Germany; 2011:845-850.
50. Bao X, Li H, Zhao G, Chang L, Zhou J, Li Y. Efficient clustering v2v routing based on PSO in VANETs. Measur. 2020;152:107306.
51. Ren M, Khoukhi L, Labiodt H, Zhangt J, Veque V. A mobility-based scheme for dynamic clustering in vehicular ad-hoc networks
(VANETs). V Commun. 2017;9:233-241.
SENOUCI ET AL. 21 of 21

52. Morales MMC, Hong CS, Bang Y. An adaptable mobility-aware clustering algorithm in vehicular networks. In: 2011 13th Asia-Pacific
Network Operations and Management Symposium. Taipei, Taiwan; 2011:1-6.
53. Souza E, Nikolaidis I, Gburzynski P. A new aggregate local mobility (ALM) clustering algorithm for VANETs. In: 2010 IEEE International
Conference on Communications. Cape Town, South Africa; 2010:1-5.
54. Sayed HE, Zeadally S, Puthal D. Design and evaluation of a novel hierarchical trust assessment approach for vehicular networks. V
Commun. 2020;24:100227.
55. Dutta A, Elhoseny M, Dahiya V, Shankar D. An efficient hierarchical clustering protocol for multihop internet of vehicles communication.
Transact Emerg Telecommu Technol. 2019:e3690.
56. Xie J, Li C. Weight clustering based TDMA-MAC scheme in VANET. Automatika. 2016;57:252-260.
57. Hadded M, Muhlethaler P, Laouiti A, Azzouz Saidane L. A novel angle-based clustering algorithm for vehicular ad hoc networks. Vehicular
Ad-Hoc Networks for Smart Cities. Singapore: Springer Singapore; 2017:27-38.
58. Hadded M, Zagrouba R, Laouiti A, Muhlethaler P, Saidane LA. A multi-objective genetic algorithm-based adaptive weighted clustering
protocol in VANET. In: 2015 IEEE Congress on Evolutionary Computation (CEC). Sendai, Japan; 2015:994-1002.
59. Khan MF, Aadil F, Maqsood M, Bukhari SHR, Hussain M, Nam Y. Moth flame clustering algorithm for internet of vehicle (MFCA-IOV).
IEEE Acc. 2019;7:11,613-11,629.
60. Bello Tambawal A, Md Noor R, Salleh R, Chembe C, Oche M. Enhanced weight-based clustering algorithm to provide reliable delivery
for vanet safety applications. PLOS ONE. 2019;14(4):1-19. https://doi.org/10.1371/journal.pone.0214664
61. Farhan A, Javid T, Albert Z. Lica: Robust localization using cluster analysis to improve gps coordinates. In: Proceedings of the First ACM
International Symposium on Design and Analysis of Intelligent Vehicular Networks and Applications. New York, NY, USA; 2011:39-46.
62. Sethi V, Chand N. A destination based routing protocol for context based clusters in vanet. Commun Netw. 2017;9:179-191.
63. Aravindhan K, Suresh Gnana Dhas C. Destination-aware context-based routing protocol with hybrid soft computing cluster algorithm for
vanet. Soft Computing; 2018.
64. Tian D, Wang Y, Lu G, Yu G. A vanets routing algorithm based on Euclidean distance clustering. 2010 2nd International Conference on
Future Computer and Communication, Vol 1. Myanmar, Burma: SCIence and Engineering Institute, USA; 2010:V1–183-V1–187.
65. Azat B, Hong T. Destination based stable clustering algorithm and routing for vanet. J Comput Commu. 2020;08:28-44.
66. Ahwazi A. Z., NooriMehr M. Mosic: Mobility-aware single-hop clustering scheme for vehicular ad hoc networks on highways. Intern J
Adv Comput Sci Appl. 2016;7(9):424-431.
67. Liang B, Haas ZJ. Predictive distance-based mobility management for PCS networks. IEEE INFOCOM '99. Conference on Computer Com-
munications. Proceedings. Eighteenth Annual Joint Conference of the IEEE Computer and Communications Societies. The Future is Now (Cat.
No.99CH36320), Vol 3. 1999:1377-1384.
68. Caballero-Gil C, Caballero-Gil P, Molina-Gil J. Self-organized clustering architecture for vehicular ad hoc networks. Inter J Distrib Sens
Netw. 2015;11(8):1-12.
69. Goonewardene RT, Ali FH, Stipidis E. Robust mobility adaptive clustering scheme with support for geographic routing for vehicular ad
hoc networks. IET Intell Transp Syst. 2009;3(2):148-158.
70. Abbas A. H., Habelalmateen M. I., Audah L, Alduais NAM. A novel intelligent cluster-head (ICH) to mitigate the handover problem of
clustering in vanets. Intern J Adv Comput Sci Appl. 2019;10(6). https://doi.org/10.14569/IJACSA.2019.0100627
71. Zhang Z, Boukerche A, Pazzi R. A novel multi-hop clustering scheme for vehicular ad-hoc networks. Proceedings of the 9th ACM
International Symposium on Mobility Management and Wireless Access. New York, NY, USA: ACM; 2011:19-26.
72. Azizian M, Cherkaoui S, Hafid A. S.. A distributed d-hop cluster formation for VANET. In: 2016 IEEE Wireless Communications and
Networking Conference. Doha, Qatar; 2016:1-6.
73. Ucar S, Ergen S. C., Ozkasap O. Multihop-cluster-based IEEE 802.11p and LTE hybrid architecture for VANET safety message dissemina-
tion. IEEE Transact V Technol. 2016;65(4):2621-2636.
74. Zhang D, Ge H, Zhang T, Cui Y, Liu X, Mao G. New multi-hop clustering algorithm for vehicular ad hoc networks. IEEE Trans Intell Trans
Sys. 2019;20(4):1517-1530.
75. Zong Y, Xu G, Jin P, Zhang Y, Chen E. Hc-ab: A new heuristic clustering algorithm based on approximate backbone. Inform Process Lett.
2011;111(17):857-863.
76. Sheu T, Lin Y. A cluster-based tdma system for inter-vehicle communications. J Inform Sci Eng. 2014;30:213-231.
77. Maslekar N, Boussedjra M, Mouzna J, Labiod H. A stable clustering algorithm for efficiency applications in VANETs. In: 2011 7th
International Wireless Communications and Mobile Computing Conference. Istanbul, Turkey; 2011:1188-1193.

How to cite this article: Senouci O, Harous S, Aliouat Z. Survey on VANET clustering algorithms: Overview,
taxonomy, challenges, and open research issues. Int J Commun Syst. 2020;e4402. https://doi.org/10.1002/dac.4402

You might also like