International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.

6, December 2024

CLUSTERING BASED ON HYBRIDIZATION OF

GENETIC ALGORITHM AND IMPROVED K-MEANS
(GA-IKM) IN AN IOT NETWORK
Moez Elarfaoui 1, Nadia Ben Azzouna 2
1
University of Tunis, ISG Tunis, SMART-LAB, 2000 Bardo, Tunisia
2
University of Tunis, ESSECT Tunis, Tunisia

ABSTRACT
The continuous development of Internet infrastructures and the evolution of digital electronics, particularly
Nano-computers, are making the Internet of Things (IoT) emergent. Despite the progress, these IoT
objects suffer from a crucial problem which is their limited power supply. IoT objects are often deployed as
an ad-hoc network. To minimize their consumption of electrical energy, clustering techniques are used. In
this paper, a centralized clustering algorithm with single-hop routing based on a genetic algorithm and
Improved k-means is proposed. The proposed approach is compared with the LEACH, K-means and OK-
means algorithms. Simulation results show that the proposed algorithm performs well in terms of network
lifetime and energy consumption.

KEYWORDS
IoT - Network-BS - Clustering - CH - LEACH - Genetic algorithm - K-means - Optimization - Energy

1. INTRODUCTION
The Internet of Things is a field with a full effervescence, it has invaded many domains. Its
applications are found in agriculture, the military, transport, and many other domains. The
number of IoT devices is continuously growing and billions of them will be linked to the Internet
[1], [2]. These equipments present serious challenges such as deployment, security, mobility, and
especially the consumption of electrical energy. Indeed, the majority of themare often battery-
powered and it is almost impossible to replace the battery in harsh environments [3]. When they
are deployed as a Wireless Sensor Network (WSN), IoT objects suffer from a lifespan limited by
the capacity of the batteries. Among the approaches used to minimize the energy consumption of
these devices and therefore the lifetime of the entire network is clustering [4]. This technique
consists of dividing all the nodes of the network into clusters. Each of these clusters is managed
by a particular node, which is often called the Cluster Head (CH). This CH takes care of the
organization of the communication inside the cluster, the collection of data from the member
nodes, and finally the sending of these packets, after aggregation, to the base station. The network
of IoT objects will therefore be reduced to a network of CH nodes. Increasing the lifetime of such
a network amounts to forming clusters that optimize energy consumption in the entire network.

Clustering techniques as well as routing techniques have a considerable influence on energy

consumption. Routing consists of finding the optimal path passing through the CH nodes to send
a packet from a node to the destination (Base Station).

DOI:10.5121/ijwmn.2024.16602 19
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
The clustering procedure can be carried out within the base station (centralized clustering) or by
cooperation between the nodes (distributed clustering). Similarly, the routing procedure, in turn,
and according to the same principle, can be centralized or distributed [5], [6], [7].

Also, the establishment of the path, for the routing of the packets, can be in mono-hop or multi-
hop. In mono-hop routing, the packets are sent directly from the CH node to the base station
without any intermediate nodes. While in multi-hop routing, the packets are emitted from a CH
node to another until the BS.

The clustering procedure is an NP-hard problem [8].As a result, meta-heuristics and artificial
intelligence techniques are often used to provide approximate solutions [9].

In this paper, the problem of centralized clustering in an IoT network with a mono-hop routing is
invoked. The purpose of the proposed solution is to maximize the lifetime of the network.
Therefore, our contribution focuses on the following points:

a) Calculating the number of clusters to be built: the number of CH nodes influences the overall
energy consumption in the network. A smaller number of CH nodes generate more
communication load between the CH and its member nodes, while a very large number
generates more traffic in the network.
b) Election of suitable CHs: The election of the CH node must take into account the amount of
energy, its distance from the base station, and the load balancing between the CH nodes.
c) Studying the BS coordinates in the area to be monitored: In many WSN studies, the choice
of position is arbitrary and not well argued [12,13,14,15]. In this work, the influence of the
base station position on the network lifetime extension is studied.

The remainder of the paper is structured as follows. Section 2presents the related works in
clustering and some kinds of hybridization. Section 3details the proposed approach then, in
Section 4, the simulation results are discussed. Finally, in the last section, the conclusion is
elaborated and different perspectives are presented

2. RELATED WORKS
The clustering procedure is aNP-problem and can encompass one or more objects
simultaneously. As a result, several techniques have been developed. These techniques depend on
the objectives and the specificity of the network nodes; homogeneous, heterogeneous, mobile,
etc. Due to the energy limitation of sensor nodes, cluster-based approaches have gained immense
research interest in the last few decades. Clustering is an effective way of conserving WSN's
energy and organizing a large number of nodes efficiently [10].

The clustering process includes some characteristics that enable the construction of a suitable
cluster: Cluster properties, cluster head properties and the cluster formation process [30].

Cluster properties are concerned with size in terms of the number of member nodes, the number
of clusters, inter-cluster and intra-cluster communication [30,31].

Cluster head properties cover mobility (mobile or stationary), its role (relay or fusion), and nature
(nodes are heterogeneous or homogeneous).As for the clustering process, cluster head selection
can be centralized or distributed. Also, several CH selection methods have been statied in the
literature. They are classified into different types: probabilistic, attribute-based, weighted
probabilistic, optimization-based, etc. [31]. The technique used can also categorize the clustering

20
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
process, which can be metaheuristic or non-metaheuristic (exact)[31]. Metaheuristic methods
consist of converting a theory inspired by nature or bio-inspired into mathematical calculations to
solve optimization problems such as Particle Swarm Optimization (PSO), Firefly Algorithm
(FA), Genetic Algorithm (GA), Ant Colony Optimization (ACO), etc.[33]. Grouping nodes into
clusters aims to optimize energy consumption and maximize network lifetime. The performance
of a clustering procedure depends on the number and distribution of CHs in the network. Since
the clusters created are profoundly affected by the CH selection method, this is an important
stepof the clustering process [32].

In the following Table.1, we present some works using a centralized clustering process and
mono-hop routing. The election of CH nodes is based on both metaheuristic and non-
metaheuristic techniques.

Table.1. some related works

Ref N° Algorithm Main idea

The fitness function is based on:
[11] GA -The distances between the node members and the CH
-The distance between the CH and the BS
The fitness function for cluster-head election is based on:
Genetic-Algorithm -The amount of residual energy,
[12]
(GAEEC) -The distance from the base station and
-The density of nodes around specific node
Fitness based on
-Sum of the distances of the nodes to the base station
[13] GA
-Distance from nodes to the CH and the distance from CH to BS
-The residual energy of nodes and the number of clusters
-Using weight of CH and weight of cluster for CH election
[14] Improved k-means -If the weight of the node is equal to the weight of the cluster, it
will be selected as a CH node
CH election uses:
-The node closest to the centroid with sufficient energy
[15] OK-K means
- A node elected as CH during the previous round, must not be
taken
Energy-saving -Centralized solution: K-mean is executed on the BS
[16] clustering algorithm -Then CH election is based on the residual energy of nodes and the
(ESCA) distance of their cluster regard to others nodes.
-Choosing initial cluster centers by applying a probabilistic
method
[17 Improved K-means
-Recalculates the sum of distances between each node and its
closest cluster center
-Using K-means to create the initial population
hybridization
-Then the genetic algorithm to create the clusters
[18] between the GA
Fitness function is defined based on intra and inter cluster
and K-means
distances and the number of alive nodes in the network
[19] Improved LEACH GA used to select the CH with more residual energy

In the previous works, the objective function is based on the amount of residual energy and the
distance between the node and BS or centroid. This distance is static (stationary nodes) and the
same node can be elected in many rounds, it results that these nodes dead quickly. Adding the
amount of residual energy as another parameter to elect CH node avoids this issue. Nevertheless,
the same node can be also elected in many rounds. The election procedure of the CH node must
be fair for all eligible nodes in order to extend their lifetimes. The frequency of election in the
previous rounds should be taken into consideration to select CH.

21
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
In many works, the start points of K-means are chosen randomly. Since K-means is sensitive to
the K starting points, it is suitable to find a solution to select these K starting points in an optimal
way. Moreover, the position of the BS is fixed arbitrarily and not justified. In the case of random
deployment of the nodes and with single hop routing, the position of BS has an impact on the
network lifetime.

3. PROPOSED APPROACH
The proposed solution consists of a hybridization of Genetic Algorithm and Improved K-Means
(GA-IKM). As K-means is sensitive to the k starting points [20], in terms of number and nature,
the main idea is to choose the k starting points in an efficient way for K-means operation.

The selection of the starting points is entrusted to a genetic algorithm while the formation of the
clusters is carried out by k-means with other improvements that take into account the constraints
specific to IoT networks. Firstly, we present the procedure of the genetic algorithm then,
secondly, we explain the improvements of k-means (3.3.3).

3.1. Network Model

The network model consists of a set of nodes, which are randomly deployed in a field. The base
station (BS) is deployed in the center of the field.

The solution is centralized and it is run in the Base Station (BS). Therefore, it is necessary to set
the following assumptions:

- All nodes are homogeneous. So they have the same technical characteristics; computing
capacity, storage capacity and amount of energy.
- The nodes are stationary (not mobile).
- The base station is able to determine the geographic coordinates of the nodes.
- A node can change its transmission range according to its needs.
- All nodes use the same packet size

3.2. Energy Dissipation Model

The architecture of an IoT object is essentially based on three parts. Sensing module, processing
module and communication module (radio module) [21]. The communication module is the
greediest in terms of energy consumption [22]. It depends on the structure of the network and the
routing protocols. In the rest of this work, we focus on the communication module. There are
different radio energy models in the literature [34]. The most common energy model used is that
of the first order Fig.1.

Fig.1. Radio Energy Dissipation Model [23]

During a communication procedure, a node can play the role of a transmitter or a receiver.

22
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
The consumed energy by a node N to send a packet of L bits over a distance d is given by the
formula (1) [24].

𝐿 ∗ 𝑑4 𝑖𝑓𝑑 ≥ 𝑑0
ETX (N) = Eelec * L + Eamp * L(1) where 𝐸𝑎𝑚𝑝 = {
𝐿 ∗ 𝑑2 𝑖𝑓𝑑 < 0

Eelec : is the electronic energy required for signal processing such as filtering, modulation.

𝐸𝑓𝑠
d0: a threshold distance. 𝑑0 = √𝐸
𝑎𝑚𝑝
Efs and Eamp are the amplification energy in the free space and multi-path fading careers,
respectively.

The consumed energy by a node N to receive a packet of L bits is given by the formula (2).

ERX(N) = Eelec * L (2)

3.3. Algorithm Flow

The proposed algorithm begins with an initialization phase of the network parameters. The data
transmission mechanism from the sensor nodes to the base station takes place in the form of
rounds [35]. In our case, each round proceeds in these steps:

(i) Computing the K number, (ii) Election of the CH nodes, (iii) Creation of clusters and (iv)
Broadcasting the clustering scheme to all nodes. Once this is done, the data transmission begins.
At the end of each round the nodes communicate their amounts of residual energy to the base
station which proceeds to a new clustering scheme for the next round. Fig.2 resumes all the steps
of the solution.

Fig .2. Flowchart of GA-IKM

23
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
3.4. Computing of the Number K

The number k must be chosen in a rigorous way in the goal to minimise the energy consumption
in the network. Several techniques were used like rule of thumb, information Criterion approach,
the Silhouette, the elbow method, Cross-validation, and Information Theoretic Approach [25,
26].

Most of these methods need to execute the clustering algorithm several times to fix the adequate
value of K. In this study, we launch GA with different values of K starting from 2, and we note
the K that minimizes the cost function.

This function corresponds to the total energy consumed by the network and it is calculated as:

k Si
(3)
𝐄𝐓𝐨𝐭𝐚𝐥 = ∑[ E(CHi ) + ∑ E(Nj ) ]
i=1 j=1

Where:

E (CHi) is the consumed energy by ith CH and

E (Nj) is the consumed energy by the jthnode, closest to CHi.
E (Nj) = ETX(Nj)(given by formula (1) in section 3.2)
E (CHi) = ERX(CHi) + ETX (CHi) (4)
ERX (CHi) = (ERX +EDA*L) *Si (5)
EDA is the energy consumed in data aggregation
Si: the number of the CHi’s closest nodes (taking into account the distance)

This objective function is very important because it improves the behaviour of K-means. This
later in sensitive to starting points that are often chosen randomly. Using GA with this objective
function allows to better orient k-means when calculating the centroids.

In addition, this function is used at the first round of clustering to compute the initial K optimal
number of clusters. For the rest of rounds, the number K is calculated by the projection of the
number of alive nodes on the interval [0,Kinitial].

For example, if in the first round we have found an optimal value for K which is 9 for 100 alive
nodes, in this case Kinitial is 9.For subsequent rounds, K is chosen by resizing the number of alive
nodes over the interval [0, Kinitial].The resizing data between a set of arbitrary values [a, b] is
given by formula (6)
(CurrentValue − Minimum) ∗ (b − a)
Transformed_Value = Round (a + )
(Maximum − Minimum)
In our case[a,b] = [0, Kinitial] (6)

(Ni − 1) ∗ (K initial − 0)
K initial = Round (0 + ) where:
(N − 1)

Ki: the value of K to be computed at roundi.

Ni: Number of alive nodes at round i.
N: The initial number of alive nodes (Maximum number of alive nodes).
The final number of alive nodes is 1 (Minimum number of alive nodes).

24
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
Example:

If we have N=100 then, the initial number of alive nodes Kinitial=9

(80-1)*(9-0)
If Ni= 80, the number of alive nodes at round i, then Ki =Round (0+ (100-1) ) = 7
If Ki=0, then the alive nodes communicate directly with the BS.

3.5. Determination of Starting Points for K-Means

The Determination of starting points for k-means is performed by the GA[27].Starting points
correspond to some k points chosen randomly as the initial centroids for k-means.

An individual from the initial population represents a potential solution [28]. This individual is
represented by one chromosome that has k genes. Each gene represents a node among the alive
nodes. The chromosomes are built in a random way and gene coding is numeric and uses decimal
format. Each gene represents an id of a node in the network. Fig.3 gives an example of a
chromosome with size = 10

20 3 14 16 100 43 80 53 94 9

Node 20Node3 . . . Node 9

Fig .3. Representation of a chromosome

3.5.1. Evaluation

The population is evaluated to take only the individuals which have the best value of the fitness
function. The way to calculate this function depends on the amount of energy that is consumed
during a round (Etotal in 3.2). The energy consumption is much higher for CH nodes because this
type of nodes has more load than its members. The CH collects data from its members, operates
the aggregation and finally transmits the messages to the base station. Energy consumption
depends also on the number of member nodes. To optimize the consumption, we have to build
clusters that are balanced in terms of the number of member nodes. As a result, the objective
function depends on the energy consumed by a CH node and the balancing of the sizes of
clusters. To measure the balance, the standard deviation of CH neighbours (CHngb) is used.
σ(CHngb) = 0 if all CHs have the same number of neighbours.

Fitness= W1* ETotal+ W2*σ(CHngb). (7)

W1and W2 represent weights.

The total energy consumed by the network (ETotal) is given by the formula (3) in 3.3.1

3.5.2. Selection

The selection operator is used to select the parents that will participate in the crossover step. In
our case, we used the Truncation selection because it is simple to implement and it conserves the
best parents for the crossover.

25
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
3.5.3. Crossover and Mutation

The crossover is performed by dividing the chromosomes into sub-sequences. The crossover
operator can be one-point crossover, Two-point crossover, K-point or uniform crossover [36].
The first operator is the common used. It allows conserving the most part of the genetic
information. The mutation consists of randomly choosing some genes, which will be replaced by
others, adding or deleting them. The rate depends on the size of the population, the number of
genes or the type of the problem. Typically, in the literature, it is chosen at 10%.

Fig .4the procedure of crossover

Once the crossover is done, we randomly take genes from each descendant to be muted with
respect to the mutation rate. Also, the mutation should not result in duplicate genes in the
chromosome fig .5.

Fig .5the procedure of mutation

Example: For the descendant 2, if we the gene 53 to be muted and the result gives 75. In this
case we obtain two occurrences of the gene 75. So the gene taken randomly from the set of N
nodes except those existing in the chromosome. The output of the genetic algorithm is a vector of
K nodes, which will serve as starting centroids for k-means to perform the procedure of
clustering.

3.6. Clusters Building Throw K-Means

Once we have selected the starting points (nodes) by GA, we form the clusters throw k-means
algorithm. These starting points will guide the algorithm to work efficiently. Once k-means was
executed, we proceed to the selection of the final CHs.

26
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
CH nodes’ election in k-means algorithm

In the basic version of k-Means [24], the CHs election is based on the Euclidean distance. Then
we choose the node closest to the centroid of the cluster. In our study, the CH node is elected by
taking into consideration some criteria witch are:

• Distance from the candidate to the BS, the amount of residual energy and the frequency of the
election as CH in the previous rounds.
-Let A be the set formed by the p nodes in the cluster. A= {n1n2, n3, . . ., np}, p is the number of
member nodes in the cluster.

-Let C be a subset of candidate nodes likely to be chosen as CH node. C= {c1,c2…cq}, C ⊂ A. q

<= p

-The set C of candidates is formed by the nodes, which have (i) the maximum amount of residual
energy, (ii) the lowest election frequencies and (iii) The lowest distances to BS.

For this purpose, each node is assigned a frequency of participation as CH per round.
Initially, all the nodes have the same frequency (zero) and the same residual energy (100%).

At each round, the frequency of participation for each node is updated as well as the amount of
residual energy. At the end of each round, we recalculate C as follows.

-Let SE be the set of nodes which have amount of residual energy greater than a threshold. The
threshold corresponds to the amount of energy required to transfer a packet to the BS.

-Let SF be the set of nodes with the minimum of election frequency in the previous rounds.

-Let SD be the set of nodes with the minimum distances from BS.

So, C = SE ⋂ SF ⋂ SD. If length(C) >= 1 we take a randomly node. Otherwise we choose the
node that has the max of residual energy.

After the formation of the clusters some member nodes are closest to the BS than to their CH
nodesfig.6. In this case they will be more energy dissipation because the packets will be
transmitted from the member node to the CH and then from the CH to BS. This value exceeds
that which separates a member node from the BS from the node member to the BS. Then nodes
close to BS can communicate directly with itfig.7. In this case the BS will be considered as an
additional CH.

27
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
Table.2. Pseudo code of Genetic Algorithm and Improved K-means

Genetic Algorithm IKM

Input : Input: Network, K, N
k , Network , N /* Total node number*/ Output : K_Clusters /* matrix of K Clusters*/
Output: StartingPoints /*vector of starting Begin
points*/ CHs = [1..k] of nodes
Begin StartingPoints=GA(K, Network, N)
MaxIteration =100, SizePop = 40 , It = 0 [Centroids, Members]=K- Means(K, Nework,
Population = [1..SizePop] of Chromosome StartingPoints)
Population = CreateRandomPop(K, Network, N) CHs= CH_Elections(Centroids, Members,
Population = Evaluate (Population) Network)
Repeat ClustersModel= ClustersFormation(CHs,
NewPopulation=Crossover(Population) Network, N)
Mutation (NewPopulation, RateMut) End
NewPopulation = Evaluate (NewPopulation)
Population=SelectBest(NewPopulation,50%)
It=It+1
Until It=MaxIeration
StartingPoints=Population
End

Fig.6. Nodes closest to BS

Fig.7. BS as a cluster head (CH)

28
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
Therefore, our contribution can be resumed in three steps.

Step 1: We use a genetic algorithm to select the k starting points for k-means.
Step 2: The election of the CHs is based on the amount of residual energy of the node, the
distance to BS of this node and its frequency of election in the previous rounds.
Step 3: We include the BS as a supplementary CH.

4. SIMULATION RESULTS AND DISCUSSION

4.1. Simulation

The simulation is performed on a set of 100 nodes which are randomly deployed on a square
surface of 100x100m. All nodes are homogeneous. In addition, the phenomena of disturbance
and signal interference are not taken into account. The evaluation criteria in our simulation focus
on:

FDN (First dead node): Corresponds to the round number at which the first node is exhausted.

LDN (Last Dead Node): Is the round number at which all nodes are exhausted and therefore the
whole network is dead.

Residual Energy: The amount of residual energy in the whole network.

PDR (Packet Delivery Rate): This factor is very important because it tells us about the robustness
of the network and the guarantee of packet reception. It is defined as the ratio of the number of
received packets to the number of packets sent [29].

Our proposed solution is compared to three algorithms that are; LEACH, K-means [24] and OK-
means [15]. Leach is the reference algorithm for the study of WSN, K-means is very used for its
convergence speed and OK-mean witch still recent and give interesting results in terms of
extending the network lifetime.

4.1.1. Simulation Environment

The simulation environment used in our work is MATLAB 2020 from Math works.
The Simulation parameters for the genetic algorithm and network are mentioned in Table.3.

Table.3. Parameters of simulation

G.A parameters value Network parameters Value

Initial population size 40 Region size 100 x 100 m
Maximum number of iterations 100 Number of nodes 100
Mutation rate 0.1 BS position (50,50)
Data Packet size 4000 Bytes
Control Packet size 100 Bytes
Initial Energy for a node 0.5 j
Transmitter/Receiver electronics 50 nj
Efs 10pj
Eamp 0.0013pj
EDA 5nj
d0 87m
Probability of CH election ( LEACH) 0.5

29
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
In the literature of genetic algorithm, there is not one initialization method that works correctly
on all problems. The choice of these values is empirical to have a reasonable running time with
an acceptable result.

4.2. Results and Discussion

The simulation results given below use single hop routing and the same energy dissipation
function presented by equation (3) in section number (3.3.1).
To better understand the advantages of our proposal we present the four use cases below.

• Case1: (IKM) we executed k-means with the improvement of the CH election that is explained
in section 3.3.1. In this case the FDN took place at round 1546.
• Case 2: We used GA to select the k starting points for K-means. Here we got the FDN at the
round 1875.
• Case 3: We launch IKM with taking into consideration the BS as a CH node. The FDN took
place at round 1900.
• Case 4: In this case we combine all possibilities. We used the GA to select the starting k points
for IKM and we used BS as a CH node. So the FDN occurred at round 1988. Table.4resumesthe
results of these use cases.

Since the positions of the nodes are chosen randomly and we have the random aspect of GA,
Referring to the work of J.Darrcet al.[29] and with numerous GA executions, a number of 32 is
sufficient to prove the superiority of our solution over the others.

Table.4.The four improved versions of k-means

Use cases FDN LDN

Improved K-Means(IKM) 1546 2057
GA-IKM 1875 2054
IKM with BS as CH 1900 2340
GA-IKM with BS as CH 1988 2354

We compared our proposal to the k-means, LEACH and Ok-means by considering the FDN,
LDN, Total residual energy and PDR metrics.

 First Dead Nodes (FDN)

Fig.9 indicates the number of dead nodes per round. This parameter represents the steady phase
of the network. The more this value is delayed the more is efficient the clustering algorithm.

30
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024

Fig.8. First dead nodes

Table.5 shows clearly the FDN and LDN for the four algorithms
Table.5. FDN and LDN

Algorithms LEACH K-means OK-means GA-IKM

FDN 1413 1288 1489 1988
LDN 1554 2294 2287 2354

Generally, the nodes that will be exhausted first are the CH nodes because they are the ones that
have more loads and therefore consume more energy. Therefore, the election of the appropriate
CH is a crucial operation. With the same simulation parameters, we note that our proposed
solution gives better results compared to LEACH, K-means, OK-means and even ESCA [16].

The election of CH in LEACH is based on probability and therefore the factors influencing the
energy consumption are not taken into account in an exhaustive way. In the basic version of the
k-means algorithm, the election of the CH node is based on the minimum distance from the
centroid. For the case of OK-means, which resembles ours, the election of the CH node takes into
account the node closest to the centroid with sufficient energy and which was not selected during
the previous round. Our solution improved k-means by introducing starting points (unsing GA)
and in addition during the election of the CH nodes, the distance from the BS and the frequency
of previous elections are taken into account.

 Residual Energy

Fig.9 shows the amount of residual energy in the whole network. Here we notice again that our
proposed solution gives most improvements even with a slight improvement over OK-means.

31
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024

Fig.9. the total residual energy

 Packet Delivery Rate (PDR)

Fig10 shows the PDR generated by the four algorithms. We constant that our proposed solution
gives the best value. This is because our algorithm has the most important FDN which makes it
possible to guarantee the correct transmission and reception of the packets in the network.

Fig.10. Packet Delivery Rate

 BS position

Now we discuss the influence of the choice of the BS position on the consumption of
energy.Table.6 shows the impact of changing BS location.

Table.6. the impact of the BS position

BS position (0,0) (0,100) (50,100) (50,0) (50,50)

FDN 1815 1643 1869 1920 1988
LDN 2369 2341 2360 2389 2354

When the base station is at the end of the field, the FDN weakens rapidly and therefore the
lifetime of the network. Whereas, if the BS is placed in the center of the area to be supervised, the
FDN will be delayed. This is due to single hop routing. For large jumps the CHs will consume
more energy. The last column in Table.6 shows the importance of deploying the BS in the center.

32
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
We notice a slight deterioration in performance for the LDN but despite this deterioration, they
remain good. This is due to the random deployment of nodes. Nodes closest to the BS consume
less energy and therefore their lifespan is longer, which gives a larger LDN.

 The deployment area

Table.7shows the impact of the deployment area. By increasing the deployment area and for a
constant number of nodes, we notice a drop in the FDN value. This is because we used single-hop
routing, the distance between the CHs and the BS becomes more and more important.

Table.7. the impact of the deployment area

Number of nodes: 100, BS position: in center of area

Deployment Area FDN LDN
100 x 100 m 1988 2354
200 x 200 m 1185 2280
300 x 300 m 325 2265
400 x 400 m 168 2000

5. CONCLUSION AND PERSPECTIVES

In an IoT network, the optimization of energy consumption is a major challenge to be solved. The
clustering technique is an efficient means of optimization. In thiswork, an improvement of the k-
means algorithm by hybridization with a genetic algorithm is proposed in order to elect,
efficiently, then CH node. This suggestion gives significant results mainly for FDN which
increases the lifetime of the network but it does not adapt to large networks and it is sensitive to
the base station position. This is due to the use of a single-hop routing. Since the communication
phase is the greediest of energy consumption, in the future research we can focus on more
energy-efficient routing protocols. To deal with this issue the integration of reinforcement
learning is a current trend for optimization in IoT networks. This technique allows adaptation to
dynamic changes in the network. In addition, optimization in such an IoT network depends on
several parameters like, residual energy, distance, reliability of nodes ... etc. therefore, a multi-
objective solution will be much more interesting whether for clustering or for routing.

REFERENCES

[1] Rani, S., Ahmed, S.H. &Rastogi, R. Dynamic clustering approach based on wireless sensor
networks genetic algorithm for IoT applications. Wireless Netw 26, 2307–2316 (2020).
[2] HUSSAIN, Muhammad Zunnurainet HANAPI, Zurina Mohd. Efficient Secure Routing
Mechanisms for the Low-Powered IoT Network: A Literature Review. Electronics, 2023, vol. 12,
no 3, p. 482.
[3] Abdulzahra, Ali Mohammed Kadhim, Ali Kadhum M. Al-Qurabat, and Suha Abdul Hussein
Abdulzahra. "Optimizing energy consumption in WSN-based IoT using unequal clustering and
sleep scheduling methods." Internet of Things 22 (2023): 100765.
[4] DEL-VALLE-SOTO, Carolina, RODRÍGUEZ, Alma, et ASCENCIO-PIÑA, Cesar Rodolfo. A
survey of energy-efficient clustering routing protocols for wireless sensor networks based on
metaheuristic approaches. Artificial Intelligence Review, 2023, p. 1-72.
[5] Del-Valle-Soto, C., Rodríguez, A. &Ascencio-Piña, C.R. A survey of energy-efficient clustering
routing protocols for wireless sensor networks based on metaheuristic approaches. ArtifIntell Rev
56, 9699–9770 (2023). https://doi.org/10.1007/s10462-023-10402-w.
[6] Yuste-Delgado, Antonio-Jesus, Juan-Carlos Cuevas-Martinez, and Alicia Triviño-Cabrera. "A
distributed clustering algorithm guided by the base station to extend the lifetime of wireless sensor
networks." Sensors 20.8 (2020): 2312.
33
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
[7] Shahraki, Amin, et al. "A survey and future directions on clustering: From WSNs to IoT and
modern networking paradigms." IEEE Transactions on Network and Service Management 18.2
(2020): 2242-2274.
[8] Wohwe Sambo, Damien, et al. "Optimized clustering algorithms for large wireless sensor networks:
A review." Sensors 19.2 (2019): 322.
[9] Iwendi, C., Maddikunta, P. K. R., Gadekallu, T. R., Lakshmanna, K., Bashir, A. K., &Piran, M. J.
(2021). A metaheuristic optimization approach for energy efficiency in the IoT networks. Software:
Practice and Experience, 51(12), 2558-2571.
[10] Rostami, Ali Shokouhi, et al. Survey on clustering in heterogeneous and homogeneous wireless
sensor networks. The Journal of Supercomputing 74 (2018): 277-323.
[11] MERAH, Malha, ALIOUAT, Zibouda, HARBI, Yasmine, et al. Machine learning‐based clustering
protocols for Internet of Things networks: An overview. International Journal of Communication
Systems, 2023, p. e5487.
[12] Singh, Santar Pal, and S. C. Sharma. "Genetic-algorithm-based energy-efficient clustering
(GAEEC) for homogenous wireless sensor networks." IETE journal of research 64.5 (2018): 648-
659.
[13] Heidari, Ehsan, et al. A novel approach for clustering and routing in WSN using genetic algorithm
and equilibrium optimizer. International Journal of Communication Systems 35.10 (2022): e5148.
[14] Razzaq, Madiha, Devarani Devi Ningombam, and Seokjoo Shin. Energy efficient K-means
clustering-based routing protocol for WSN using optimal packet size. 2018 International
Conference on Information Networking (ICOIN). IEEE, 2018.
[15] El Khediri, Salim, et al. "Improved node localization using K-means clustering for Wireless Sensor
Networks." Computer Science Review 37 (2020): 100284.
[16] NEDHAM, WisalBassim et AL-QURABAT, Ali Kadhum M. An improved energy efficient
clustering protocol for wireless sensor networks. In: 2022 International Conference for Natural and
Applied Sciences (ICNAS). IEEE, 2022. p. 23-28.
[17] Ahmad, Waseem, et al. "Optimizing Energy Efficiency in Wireless Sensor Networks using
Enhanced K-Means Cluster Head Selection." International Journal of Communication Networks and
Information Security 16.3 (2024): 565-573.
[18] Bhushan, Shashi, Raju Pal, and Svetlana G. Antoshchuk. "Energy efficient clustering protocol for
heterogeneous wireless sensor network: a hybrid approach using GA and K-means." 2018 IEEE
Second International Conference on Data Stream Mining & Processing (DSMP). IEEE, 2018.
[19] Bhola, Jyoti, SurenderSoni, and Gagandeep Kaur Cheema. "Genetic algorithm based optimized
leach protocol for energy efficient wireless sensor networks." Journal of Ambient Intelligence and
Humanized Computing 11 (2020): 1281-1288.
[20] Hassan, A. A. H., et al. "Clustering approach in wireless sensor networks based on K-means:
Limitations and recommendations." Int. J. Recent Technol. Eng 7.6 (2019): 119-126.
[21] Obeid, Abdulfattah Mohammad, et al. "A survey on efficient power consumption in adaptive
wireless sensor networks." Wireless Personal Communications 101 (2018): 101-117.
[22] Singh, Jaspreet, Ranjit Kaur, and Damanpreet Singh. "A survey and taxonomy on energy
management schemes in wireless sensor networks." Journal of Systems Architecture 111 (2020):
101782.
[23] SINGH, Shashank et ANAND, Veena. Load balancing clustering and routing for IoT‐enabled
wireless sensor networks. International Journal of Network Management, 2023, vol. 33, no 5, p.
e2244.
[24] Ray, Anindita, and Debashis De. "Energy efficient clustering protocol based on K‐means
(EECPK‐means)‐midpoint algorithm for enhanced network lifetime in wireless sensor network."
IET Wireless Sensor Systems 6.6 (2016): 181-191.
[25] Marutho, Dhendra, Sunarna Hendra Handaka, and EkapranaWijaya. "The determination of cluster
number at k-mean using elbow method and purity evaluation on headline news." 2018 international
seminar on application for technology of information and communication. IEEE, 2018.
[26] Kodinariya, Trupti M., and Prashant R. Makwana. "Review on determining number of Cluster in K-
Means Clustering." International Journal 1.6 (2013): 90-95.
[27] Kramer, Oliver, and Oliver Kramer. Genetic algorithms. Springer International Publishing, 2017.
[28] Seo, Hyun-Sik, Se-Jin Oh, and Chae-Woo Lee. "Evolutionary genetic algorithm for efficient
clustering of wireless sensor networks." 2009 6th IEEE Consumer Communications and
Networking Conference. IEEE, 2009.

34
 International Journal of Wireless & Mobile Networks (IJWMN), Vol.16, No.6, December 2024
[29] Derrac, J., García, S., Molina, D., & Herrera, F. (2011). A practical tutorial on the use of
nonparametric statistical tests as a methodology for comparing evolutionary and swarm intelligence
algorithms. Swarm and Evolutionary Computation, 1(1), 3-18.
[30] BAQERI, Javad. Increase the Lifetime of Wireless Sensor Networks Using Hierarchical Clustering
with Cluster Topology Preservation. International Journal of Wireless & Mobile Networks
(IJWMN) Vol, 2021, vol. 8.
[31] RAJ, Bryan, AHMEDY, Ismail, IDRIS, Mohd Yamani Idna, et al. A survey on cluster head
selection and cluster formation methods in wireless sensor networks. Wireless Communications and
Mobile Computing, 2022, vol. 2022, p. 1-53.
[32] AL-SULAIFANIE, Adnan Ismail, AL-SULAIFANIE, Bayez Khorsheed, et BISWAS, Subir.
Recent trends in clustering algorithms for wireless sensor networks: A comprehensive review.
Computer Communications, 2022, vol. 191, p. 395-424.
[33] DEL-VALLE-SOTO, Carolina, RODRÍGUEZ, Alma, et ASCENCIO-PIÑA, Cesar Rodolfo. A
survey of energy-efficient clustering routing protocols for wireless sensor networks based on
metaheuristic approaches. Artificial Intelligence Review, 2023, p. 1-72.
[34] ROY, NiharRanjan et CHANDRA, Pravin. Energy dissipation model for wireless sensor networks:
a survey. International Journal of Information Technology, 2020, vol. 12, p. 1343-1353.
[35] Gülbaş, Gülşah, and Gürcan Çetin. "Lifetime Optimization of the LEACH Protocol in WSNs with
Simulated Annealing Algorithm." Wireless Personal Communications 132.4 (2023): 2857-2883.
[36] Alhijawi, B., Awajan, A. Algorithmes génétiques : théorie, opérateurs génétiques, solutions et
applications. Évol. Intel. (2023). https://doi.org/10.1007/s12065-023-00822-6

AUTHORS

MOEZ Elarfaoui:

-Computer science engineer and researcher in SMART LAB.

-University Teacher at the Zaghouan Higher Institute of Technological Studies,
Tunisia

Nadia Ben Azzouna:

-Engineer and Dr in computer science
-University Teacher at ESSECT Tunis, Tunisia.

35