IEEE SENSORS JOURNAL, VOL. 21, NO.

18, SEPTEMBER 15, 2021 21099

IDCT: Intelligent Data Collection Technique for

IoT-Enabled Heterogeneous Wireless Sensor
Networks in Smart Environments
Walid Osamy , Ahmed Salim , Ahmed M. Khedr , and Ahmed A. El-Sawy

Abstract —Wireless Sensor Networks (WSNs) lend them-

selves to a wide variety of applications in our daily lives,
such as environmental monitoring, safety, health-care, animal
monitoring, etc. However, one of the key issues in WSN is
energy constraints. This makes energy-conservation one of
the major keys to the efficient functioning and lifetime of WSN.
In this paper, given a network of nodes with heterogeneous
energy, our goal is to determine energy-aware disjoint domi-
nating sets (DSs) that work as data collection nodes in each
round, to improve overall WSN lifetime. In order to accomplish
this goal, we propose an intelligent data collection technique
with two phases, the collector nodes selection, and the data
gathering path formation and collection phases. In the collec-
tor nodes selection phase, an energy-aware algorithm based on swarm intelligence is proposed to construct disjoint
dominating sets that work as collector nodes in each round. In the data gathering path formation and collection phase,
data gathering path is determined for achieving maximal data collection efficiency and reduced energy consumption. The
efficiency of our proposed technique is proved mathematically and through simulations.
Index Terms — Heterogeneous wireless sensor networks (HWSNs), the Internet of Things (IoT), the IoT-enabled WSN,
bees based algorithm, network lifetime, dominating set, energy consumption, data collection.

I. I NTRODUCTION other using a multi-hop scheme to deliver the accumulated data

to the sink node. In contrast to the normal nodes deployed for
A WIRELESS Sensor Network (WSN) comprises of a set
of sensor nodes that are used to obtain relevant data
from the environment in accordance with the application
data collection, sink nodes are considered to have high power
resources, strong analytics, sufficient memory, processing and
requirements. The nodes in the WSN interconnect with each communication capabilities [1]–[5]. Real-life applications of
WSNs are increasing by leaps and bounds [6]. There are
Manuscript received May 29, 2021; revised July 19, 2021; accepted countless applications around us, such as home monitoring,
July 22, 2021. Date of publication July 26, 2021; date of current version agriculture environmental monitoring, health care monitor-
September 15, 2021. This work was supported by Qassim Univer-
sity through the Deanship of Scientific Research under Grant ucc-bs- ing, animal monitoring, environmental monitoring, industrial
2020-1-3-I-10127 during the academic year 1441 AH/2020 AD. The monitoring, and so on, and various types of sensors are
associate editor coordinating the review of this article and approving used to accomplish these goals. The deployment of nodes
it for publication was Dr. Cheng-Sheng Huang. (Corresponding author:
Walid Osamy.) depends on the specific application requirements, and are
Walid Osamy is with the Department of Applied Natural Science, required to operate in harsh and hostile environment. Even so,
College of Community, Qassim University, Unaizah 51911, Saudi Arabia, WSNs experience a number of technical difficulties, including
and also with the Computer Science Department, Faculty of Computers
and Artificial Intelligence, Benha University, Benha 13511, Egypt (e-mail: limited battery power, reduced transmission range, restricted
[email protected]; [email protected]). storage capacity and computational limitations that needs to
Ahmed Salim is with the Department of Computer Science, College be addressed.
of Sciences and Arts, Qassim University, Buraidah, Al-Mithnab 51931,
Saudi Arabia, and also with the Mathematics Department, Zagazig WSNs may be classified as heterogeneous or homogeneous,
University, Zagazig 44519, Egypt (e-mail: [email protected]). based on their nature. The cost incurred in increasing the
Ahmed M. Khedr is with the Computer Science Department, University node’s energy is much lower than that of deploying extra
of Sharjah, Sharjah 27272, United Arab Emirates, and also with the
Mathematics Department, Zagazig University, Zagazig 44519, Egypt nodes of homogeneous energy. This indicates the effect of
(e-mail: [email protected]). energy heterogeneity in WSN, but at the same time it is very
Ahmed A. El-Sawy is with the Computer Science Department, Faculty challenging in heterogeneous WSNs (HWSNs) to save battery
of Computers and Artificial Intelligence, Benha University, Benha 13511,
Egypt (e-mail: [email protected]). power and lifetime, and to maintain various quality of service
Digital Object Identifier 10.1109/JSEN.2021.3100339 parameters. Limitation in the energy of sensor nodes leads
1558-1748 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
21100 IEEE SENSORS JOURNAL, VOL. 21, NO. 18, SEPTEMBER 15, 2021

them to survivability problems and face failures. One of the tem tuning [47], [48]. The Bees algorithm does not make
crucial issues in a WSN is to enhance energy efficiency, which any assumptions such as the continuity of the search space
can have a significant impact on the stability and lifetime of and requires limited problem domain knowledge beyond the
the WSN. Moreover, the ability of the network to withstand desired solution performance. The range of applicability of the
various challenges and failures in order to meet the targeted Bees algorithm to the field of engineering encompasses a large
tasks is a basic concern in a WSN [6], [7]. number of optimization tasks. The Bees algorithm consists of
Various methods have been presented in the literature to exploitation and exploration, the two key elements of optimiza-
preserve energy and extend network lifetime. Sensing and tion techniques that help to find the global optimum in opti-
communication are the major energy-consuming tasks, and mization problems [47]. Moreover, the algorithm has strength
therefore effective energy management and scheduling can in both local and global searches and it can be efficiently
help to extend the service life of the WSN. One of the used for solving several multi-modal and multidimensional
generic techniques for preserving network energy is switching optimization problems. In our work, we invest the capabilities
the node’s state between awake and sleep modes [8]. As the of SI and the benefits of Bees algorithm as a swarm-based
deployment of the WSN nodes is usually in a dense manner, algorithm to form disjoint DSs that work for heterogeneous
retaining the sensing and radio communication capabilities of WSN. This is done by redesigning the Bees algorithm and its
the nodes suspended or stopped while in the sleep state can parameters and operations, to fit for proposed problem.
make the sleep/awake method an efficient solution for energy In this paper, we utilize the concept of DS to select the
conservation. Only the nodes that belong to the active set take cluster head nodes, where multiple disjoint DSs are selected
the responsibility of sensing and data collection from the target based on node energies. We aim to enhance the overall lifetime
area. The remaining nodes stay in a sleep state, characterized of a heterogeneous WSN where the nodes in the network have
by low power consumption. Switching the node’s state from heterogeneous initial energy. Proper selection of the disjoint
awake to sleep and vice versa can help in providing prolonged DSs in an efficient manner can help to improve the overall
service. WSN lifetime, where the network lifetime is the sum of the
Most of the algorithms in literature focus on finding a single lifetime of each disjoint DS. The key objective of our work
dominating set (DS) in the network. Such a set can function is to consider energy as the main factor while forming the
as a virtual backbone of a WSN and helps to structure the DS of cluster heads, and to provide a swarm-based algorithm
WSN in a hierarchical manner for use in data routing and (Bees algorithm-based method) with an energy-aware fitness
communication, e.g., [20]–[22], [32], [33], [37], [40], [42], function, to enable the selection of sets that can maximize the
[46]. Moreover, the concept of DS is integrated with clustering WSN lifetime.
in WSNs for cluster head operations [34], [35], [41], [45] or The following are the main contributions of the paper:
inside each cluster [36]. Further improvements are made by • We address the issue of energy conservation in hetero-
considering the DSs problem as a domatic partition problem geneous WSN by proposing energy-aware disjoint DS
where multiple disjoint DSs are determined, e.g., [15]–[17], algorithms.
[23]. Although all the mentioned above algorithms succeeded • We present a novel method based on the Bees algorithm,
to solve the issue of finding the DS, none of them considered where the energy of the formed sets and the lifetime of
the node’s energy during the creation or selection of DSs. the network are considered in the fitness function of the
Moreover, the existing algorithms for multiple disjoint DSs do bees algorithm, for selecting sets that can enhance WSN
not always produce good results for sensors having different lifetime.
initial energy. Also, reliance on single virtual backbone may • The proposed technique uses node’s energy to effectively
cause problem in data transmission if node failure occurs, and achieve load balancing during data collection and to
create unbalance between nodes. Motivated by this, we utilized determine the path from the collector node to Base station
the DS concept to select the set of collectors, where multiple (BS), which leads to an increase in the energy efficiency
disjoint DSs are selected considering the heterogeneity of node of IoT-enabled WSNs.
energies. We provide a dynamic method of selecting collection • Provide extensive simulation results to verify that the pro-
paths for collector nodes to achieve balance between nodes. posed algorithm can enhance WSN lifetime and exceed
Several Swarm Intelligence (SI) methods have been used other related schemes with reference to energy efficiency
to address WSN issues to bring efficient optimization solu- and WSN lifetime.
tion [50], [51]. SI is employed for many challenges in WSN
The rest of this paper is structured as follows: Section II
such as network lifetime, node deployment, connectivity, cov-
presents the review of related works. The proposed approach
erage, scheduling, localization, clustering, data gathering [49],
is described in Section III. The results of simulation and the
[50]. SI techniques have also been applied to the applications
comparison with existing schemes are provided in Section IV.
of IoT system and cloud computing, to facilitate the solving
Section V concludes our work.
of multi-objective optimization problems [49]. One of such
SI algorithm is the Bees algorithm. Different versions of
the Bees Algorithm have been used to solve various engi- II. R ELATED W ORKS
neering problems, such as pattern classifier training, dynamic Different methods have been reported in the literature to
control problems, machine shop scheduling, robotic swarm address the power constraints in IoT networks based on various
coordination, non-linear model identification and control sys- factors affecting energy consumption. Numerous works have
OSAMY et al.: IDCT: INTELLIGENT DATA COLLECTION TECHNIQUE 21101

found that the implementation of a sleep/awake schedule is additional energy of nodes and proximity to the BS. CDS-
effective for conserving the nodes’ energy [6], [8]. However, LEACH [35] utilized connected DS concept as a metric for
the representation of WSN by finding the minimum DS for cluster head selection.
IoT networks is considered as an NP-hard problem. This K-connected DS (K-CDS) algorithm named CDS-FOR is
indicates that the problem of sleep/awake scheduling in WSN proposed in [45] with the goal of transmitting the data sensed
can be interpreted as an NP-hard minimum DS problem. Many by the nodes towards the sink by forming K-CDS. This ensures
research works have been introduced in order to address this that each node is connected to a minimum of k-nodes.
concern. A Balanced Energy Efficient Virtual Backbone Construction
In [14] and [15], centralized approximation algorithms were (BEE-VBC) algorithm is proposed in [46]. BEE-VBC employs
proposed. However, centralized approximation algorithms are the well-known heuristic Genetic Algorithm (GA) and con-
not convenient solutions for large-scale WSNs [15]. Thus, sider multiple factors such as energy, coverage of neighbors,
the work in [16]–[18] utilized the properties of distributed reliability, data rate, and traffic deviation in the design of
algorithms in order to create DSs. fitness function. The maximum fitness nodes are utilized for
In order to create small Connected DSs (CDS), [19] pro- CDS construction to improve the quality of the solution and
posed a degree-based three-stage centralized greedy approxi- to ensure steady connectivity among the dominator nodes, and
mation algorithm. The authors stated that the method provided form the path towards the BS.
better results compared to traditional algorithms, in terms of In [37], a method called ADCOCDS is suggested for
construction cost. The work proposed in [7] increased the reducing transmission delay and network power consumption.
CDS connectivity by using a non-trivial potential function. ADCOCDS creates a backbone network. The nodes in the
The presented greedy strategy is based on a 2-connected backbone network run in a duty cycle-based mode, while
graph decomposition (called Tutte’s decomposition) to form others conserve power by switching off the radio if no data
3-connected components. needs to be sent. If the occurrence of an event is detected by
The work in [10] proposed a clustering approach based on a dominate node, it will turn on its radio and transmit the data
DS, known as EBDSC, to improve network performance. The to the corresponding dominator.
work in [11] proposed a distributed scheme for determining An intelligent CDS-based backbone design scheme is pre-
the nodes to form a DS in a WSN. sented in [40]. A distributed approach is adopted to design the
The authors of [20] and [21] have proved the importance WSN structure, with nodes energy value serving as a major
of virtual backbone (VB) in terms of increasing network parameter. Thus a connected DS (CDS) is constructed in a
scalability and efficiency. As the connectivity offered by a distributed fashion to create a virtual backbone.
connected DS allows for the transmission of data and com- Work in [41] proposed a weighted clustering algorithm for
munication between the nodes in the WSN, such a set can WSNs based on connected DSs. The backbone structure’s
function as a VB of a WSN (VBN). Moreover, this helps to elements are chosen according to the (k, T) - CDS where,
structure the WSN in a hierarchical manner for use in data k is the backbone nodes count and T is the hop distance.
routing and communication. Connected DSs in the context of The method determines the elements of the DS one by one,
energy harvesting (CDSEH), which is based on the concept selecting the most appropriate sensor node as a member of the
of VBN, have been proposed in [22]. They applied both DS based on its weight. In multi-hop scenarios, the (k, T) -
centralized as well as distributed schemes to solve the issue. CDS gives CH redundancy.
In addition, the work proposed in [22] verified that the problem In [42], authors proposed an optimized region based- data
discussed in CDSEH is NP-complete and thus they suggested routing approach (AORED) for WSNs. In AORED, CDS
a hybrid algorithm to address it. In [23], the DS problem is was applied to construct the VBN of communicating nodes.
converted into a domatic partition problem to provide further In EBDSC scheme of [44], each node calculates its lifetime.
improvements by partitioning the WSN nodes into disjoint DS. If a node has the maximum lifetime amongst the neighbors,
Although all the aforementioned algorithms succeeded to it is selected as a candidate CH. Its decision to become a final
solve the issue of finding the minimum DS, none of them CH is according to a probability determined by its uncovered
considered the node’s energy during the creation or selection neighbors.
of CDS. Thus, the work proposed in [9] used different local The work in [43] introduced an energy constrained mini-
search algorithms, while considering the nodes varying initial mum connected DS approach and used energy constraints to
energy to create CDS. However, the complexity of the pro- design the task of determining CH optimally. The authors
posed algorithm is considered as its main drawback. Motivated of DGA-EBCDS [33] suggested that by constructing an
by this, we utilized the DSs concept to select the sets of energy-balanced CDS for data acquisition, the network life-
collectors, where multiple disjoint DSs are selected based on time could be extended.
node energies. By modifying the basic LEACH, LEACH-DS-ACO is pro-
In the context of employing DS for data collection and posed in [34]. In this work, DS formation is applied to each
routing in WSNs, the work in [36] introduced Fuzzy-DS-ACO. cluster to construct its DS. Then a cluster chain is constructed
Fuzzy-DS-ACO divides sensor nodes into clusters using Fuzzy using ACO, and leader for the chain is chosen based on
C-Means. For each cluster, a DS is identified, and the cluster remaining energy and distance to the BS. In [32] authors
chain (CC) is constructed using ACO. Using the Fuzzy Logic proposed a Hierarchically Diffused CDS (HDCDS)-based
system, a chain leader (CL) for each CC is selected based on Compressive Data Gathering (CDG) scheme. A CDS was built
21102 IEEE SENSORS JOURNAL, VOL. 21, NO. 18, SEPTEMBER 15, 2021

as the network’s backbone in HDCDS. The remaining nodes

are linked to the backbone. The traditional data collection
scheme is used by backbone nodes to gather data from the
remaining non-backbone nodes. Moreover, CDG is utilized
for transmitting data from the backbone to the BS.

III. P ROPOSED I NTELLIGENT DATA C OLLECTION

T ECHNIQUE
In this section, we propose our intelligent data collection
technique (IDCT). First, we explain the system model, and
then the working phases of the proposed technique are pro-
vided. Fig. 1. Heterogeneous WSN graph example.

A. System Model
The proposed IDCT technique operates in two phases:
In our proposed model, WSN is depicted as a graph G =
• Bees based collector nodes selection (BBCNS), and
(V, E). V and E stand for the set of deployed WSN nodes and
• Data gathering path formation and collection.
the communication links between nodes, respectively. There
exist an edge between two nodes p and q ∈ V if p lies In the first phase, we find energy-aware disjoint DSs that work
within the communication range (CR) of q. We follow the as collector nodes for each round based on the adapted Bees
assumption that the links between nodes in the WSN are algorithm in which the energy of the formed sets and the WSN
symmetric, i.e., all the nodes in the WSN possess identical lifetime are considered while devising the fitness of the Bees
communication range. The sensor nodes are battery-powered algorithm. We find the data gathering path to send data from
and have heterogeneous initial energy. The power consumption collector nodes to the BS in the second phase. Algorithm 1
rate is considered constant for all the nodes. presents the pseudo-code for IDCT.
A set D ⊂ V is a DS if ∀v ∈ V \D has a neighbor in D.
In the concept of a DS, the nodes assigned for accomplishing Algorithm 1 Intelligent Data Collection Technique (IDCT)
the tasks on behalf of their neighboring nodes constitute the 1: Input: WSN, Bees algorithm parameters.
DS, and can be used to prolong WSN lifetime. DS with 2: Output: Collecting Path for each node.
a smaller size can perform effectively in terms of power 3: DSs = DS1 , DS2 , . . . , DSk : set of dominated sets.
efficiency since more nodes can be made to sleep mode. 4: k: total number of formed DSs.
However, the major drawback of this scheme is the lack of 5: {Generate dominated sets} //Phase-I
balance in energy distribution among the nodes. This problem 6: Start Bees Based Collector Nodes Selection (BBCNS)
can be eased by substituting an existing DS with another Phase to generate DSs.
disjoint DS in the WSN and switch the state of the nodes 7: {Data gathering path formation and collection phase}
in the previous set to sleeping mode. I.e., employing as many //Phase-II
disjoint DSs as possible and utilizing each for a particular 8: Initial Step: Form aggregation tree
amount of time, then substituting it with another, and so on. 9: for each round r do
The node with minimum energy in a DS determines the 10: i x = mod(r, k) {determine the index of current DS}
lifetime of that set, which is obtained by dividing the node 11: {get DS for current round}
energy by the rate at which the energy is consumed. Our target 12: if i x = 0 then
is to enhance the entire lifetime of heterogeneous WSN. This 13: curr ent DS = DSs(i x)
can be achieved by selecting the disjoint DSs in an efficient 14: else
manner such that the total network lifetime is improved, where 15: curr ent DS = DSs(k)
the overall WSN lifetime is the sum of the lifetime of each of 16: end if
the disjoint DS. 17: Get parent node for each collector node ci ∈ curr ent DS
For clarification, consider the example of a WSN graph by calling algorithm 3.
consisting of seven nodes with heterogeneous energy shown in 18: Each ci forward data to selected parent node toward BS.
Figure 1. From the network graph, we can have two disjoint 19: end for
DSs D1 = [1, 2, 5] and D2 = [3, 4, 6]. In D1 , node 5 has min-
imum energy, while node 4 in D2 has minimum energy. There-
fore, if we consider the energy consumption rate as constant
and equal to 0.05, then the lifetime of D1 is 0.03/0.05 = 0.6 B. Bees Based Collector Nodes Selection (BBCNS)
and the lifetime of D2 is 0.18/0.05 = 3.6. As a result, the total Phase
network lifetime is 4.2. The total network lifetime can be Bees algorithm [12] is one of the popular meta-heuristic
optimized and enhanced if the DSs are D1 = [2, 4, 5] and optimization schemes because of its efficiency in evaluating
D2 = [1, 3, 6]. In this case, the overall WSN lifetime will be global results to a variety of problems. This algorithm reflects
(0.03/0.05) + (0.76/0.05) = 0.6 + 15.2 = 15.8. honey bees’ natural behavior in determining the best site
OSAMY et al.: IDCT: INTELLIGENT DATA COLLECTION TECHNIQUE 21103

among several sites in terms of both speed and accuracy. This

is similar to the search for the optimal solution in an optimiza-
tion process. Algorithm 2 gives the details of the basic bees
algorithm. Here, we adapt the Bees algorithm to optimize and
form disjoint energy-aware DSs for heterogeneous WSN that Fig. 2. Representation of bees.
works as collector nodes by redesigning the Bees algorithm
and its parameters and operations, to fit for proposed problem.
The above steps will be repeated to generate many solutions
This is explained as follows:
for use to initialize the population. Each solution will be
evaluated using the fitness function.
Algorithm 2 Bees Algorithm For clarification, the algorithm execution using the network
1: Input: Bees Algorithm parameters and population. shown in Figure 1 is given. Initially, x = {0, 0, 0, 0, 0, 0, 0},
2: Output: Best Solution (best value of decision variables). the number of DSs k = 1 and the temporary value
3: Randomly initialize population. v = −(k − 1) = 0. In the first iteration, find the
4: Calculate the fitness of the population. set of indices i x for the items in x that equals v,
5: while stopping condition not met do i.e., i x = {1, 2, 3, 4, 5, 6, 7}. Let r = 6 be the selected
6: Highest fitness values are specified as “selected sites” for random item from i x. Set x(r ) = k and change the value
neighborhood search of each neighbor of node 6 (nodes 2 and 7) from v to −k.
7: Find the size of the neighborhood around the selected The updated x will be x = {0, −1, 0, 0, 0, 1, −1} and the
sites. updated i x = {1, 3, 4, 5, }. The next step in this iteration is
8: Enroll bees for selected sites, then evaluate fitness values. to pick random id from i x say r = 5 the updated x and i x
9: Choose the best fit (fittest) bee from each neighborhood. will be x = {0, −1, −1, 0, 1, 1, −1} and i x = {1, 4} where
10: Relinquish sites that has new information. node 3 is the neighbor node for node 5. Finally, let r = 1 be
11: Perform search randomly by remaining bees and esti- the next random number. Then the updated x and i x will be
mate their fitness values. x = {1, −1, −1, −1, 1, 1, −1} and i x = {}, where neighbors
12: end while of node 1 are nodes 3 and 4.
The second iteration starts by setting k = 2 and v = −1.
First, find the set of indices i x for the items in x that equal to
1) Bee Representation and Population Initialization Step: v, i.e., i x = {2, 3, 4, 7} and let r = 3 be the selected random
Each bee position represents a solution for the given problem, item from i x. As before, when this iteration ends, we will
which contains a collection of disjoint DSs. Figure 2 gives have x = {1, −2, 2, 2, 1, 1, 2}.
an example of a solution for the graph network in Figure 1. The third iteration starts by k = 3, v = −2 and ends with
In Figure 2, the index indicates node id and the value indicates x = {1, 3, 2, 2, 1, 1, 2}, which means that the DSs are D1 =
the DS of the node. Nodes 1, 2, and 5 belong to DS number 1, {1, 5, 6}, D2 = {3, 4, 7},and D3 = {2}. Note that only valid
nodes 3, 4, and 6 belong to DS number 2, while node 7 belongs sets are considered during the evaluation.
to an invalid DS. DSs 1 and 2 are valid disjoint sets and they 2) Fitness Function: Our goal is to form energy aware
represent a solution while DS 3 is not considered during the disjoint DSs that maximize the network lifetime. Therefore,
calculation of the fitness value of that solution since it violates the fitness function F is formulated as follows:
DS condition. 
j
min(s1k .E, s2k .E, . . . , sik .E)
Generating Solution Procedure (GSP): In order to generate F = max( ), (1)
a solution the following steps will be executed: δ
k=1
1) Solution x is initialized to zero with length n, where where sik .E denotes the energy of node number i in DS Dk ⊆
n denotes the nodes count, (x = i 1 . . . i n , i j = 0, j = V , k = (1, . . . , j ), j is the count of disjoint DSs and δ is the
1 . . . n) and the count of DSs k = 1. energy consumption rate. For example, the fitness value for
2) Repeat until all the values of x are +ve. the solution in Figure 2 is 4.2 (note that set-number 3 is not
a) From x, find the indices set i x for the items with considered).
value v = −(k − 1). 3) Neighborhood Search Step: After evaluating the fitness
b) while i x is not empty value for each site, the values are ranked in decreasing order
of fitness. The m sites with the maximum fitness values are
• select random index value r from i x, then update chosen in this stage. Then, they are classified into two site
x by setting the value at index r to k types: (i) elite (e) and (ii) non-elite (m − e), depending on
• update the indices of the neighbors of the node their fitness values. This step begins by allocating elite bees
with index r and update their value from v to to search around them (the elite site will have more recruited
−k bees while the remaining selected sites will have fewer bees).
• update the set i x by removing node r and its The bees search around these positions and this is done by
neighbors. swap mutation. For this, two random DSs are selected; then
• Goto step 2a. position at each one is selected randomly and the values at
c) k = k + 1 (the set i x is empty) these positions are interchanged. Figure 3 gives an example
21104 IEEE SENSORS JOURNAL, VOL. 21, NO. 18, SEPTEMBER 15, 2021

work in [24] and [27]. When the collector node intends to

send its data, a path is established between the collector node
and the BS to deliver the data. This phase contains three steps,
initial step, parent node selection step, and data collection step.
1) Initial Step: The BS sends a PROBE message containing
hop distance to its neighbors and waits for ACK message from
neighbor nodes. Each neighbor node s compares the received
hop distance with its initial hop distance (a large number)
and updates its hop distance to the minimum hop distance
Fig. 3. Generating new solution using swap operation between sets. +1 and sets the sender in its parent’s list. Then s forwards
a PROBE message that includes its new hop distance to its
neighbors and waits for ACK from neighbors. After updating
its hop distance, if s receives a hop distance less than its hop
distance it will add the received one to its parent’s list. The
whole process will be terminated after all ACK messages are
received from children. At the end of this step, each node has
a list of parents and the best parent will be selected using the
next step.
2) Parent Node Selection Step: In this step, each collector
node selects the neighbor node from its parent list (P Li st)
Fig. 4. Generating new solution using crossover operations between to work as a relay node for the round r . The pseudo-code is
sets. detailed in Algorithm 3.
Each collector node c selects a random number for each
of this operation. The new bee position is then evaluated, and node Tr (si ), 0 ≤ Tr (si ) ≤ 1, si ∈ P Li st, which is then
the best one will be selected. compared with a threshold T (s, r ). The nodes with T (s, r )
4) Global Search Step: The global search process starts with greater than Tr (si ) serve as a relay node. Therefore, T is
(n − m) sites being randomly explored by the bees, i.e., the computed based on node density and residual energy of each
remaining (n − m) bees are subjected to crossover function si in the parent list.
with the specified rate (that is to expand the population T (s, r ) for a node si is computed as:
⎧
diversity) or random generation of new solution, where 0% ⎪ −1 −1 2
⎨ p × (1 − p × mod(r,  p )) ×  ,
rate means (n − m) is completely generated randomly while
T(s, r) = i f ω(s, r ) = 0 (2)
a 100% rate means (n − m) is generated through crossover ⎪
⎩
operation. This process is described as follows: 0, otherwise
• for each site x i in (n − m) sites 1
 = (1 − exp(−s.Er )) + (ε div )(exp(−s.Er )) (3)
• Select r from 0 to 1 (a random number). p
• If r ≤ crossover_rate, then
where, s.Er = s.E s.E
, s.E denotes the current energy, s.E o
– randomly select site y from m sites, and set z = y o

– randomly select DSs D1 and D2 from z denotes the initial energy, p = |P List1
| , |P Li st| denotes the
– randomly select index r1 from D1 and r2 from D2 size of parent list for collector node, di v represents integer
– exchange between the two sets D1 and D2 at indices division and ε is the count of successive rounds in which
r1 and r2 . a node does not work as relay node. The threshold value
increases when ε reaches the value of 1/ p, which gives the
• Else z is set to a new random solution
chance for node s to take the role of relay node. Moreover,
• x i is replaced if it has lower fitness value than z.
the value of ε is set to zero when a node is selected as a relay
Repeat the above steps to update each individual in (n − m). node. This prevents the low value for the threshold that could
Figure 4 shows an example of selected bee and the crossover occur when the nodes have a very low energy level.
operation on the sets in it to generate new bee. From Figure 4, ω(s, r ) is a function that identifies whether or not a node
we can see that the new solution contains {7}, {1, 6} and s can share the selection procedure for round r ; zero value
{2, 4, 5, 3}. The first two sets are invalid while the 3rd one indicates that s can serve as relay node while other values
is a valid set which is only considered during fitness value indicate that s cannot serve as a relay node. It is calculated as
calculations for this individual. follows:
5) Final Step: Finally, the overall positions are sorted by
1
fitness values, and the process runs until the global optimum ω(s, r ) = − mod(r, 1/ p). (4)
is discovered. p
3) Data Collection Step: Each node that is not in the set of
C. Data Gathering Path Formation and Collection Phase collector nodes selects the nearest collector node and sends
In order to determine the path of data gathering, we propose its sensed data to it. The collector node constructs a path
an intelligent data collection method that is inspired by the to the BS by determining the parent node (based on Parent
OSAMY et al.: IDCT: INTELLIGENT DATA COLLECTION TECHNIQUE 21105

Node Selection Step) that works as a relay node with the selected nodes D1 and the set of neighbors of the selected
best benefits and sends it confirm message. The selected node nodes xD1, and each node in the network is either in D1
continues the path construction by forwarding the data to its or has a neighbor in D1 . According to the definition of DS,
next node towards the BS. This process repeats until the data D1 is a DS, which contradicts the assumption that GSP does
reaches the BS and each time node with maximum benefit is not generate a solution.
selected as the next node for data delivery towards the BS. Lemma 2: BBCNS correctly generates the DSs.
Proof: We prove this through contradiction. Suppose the
Algorithm 3 getParentID: Select Parent Node to Work as claim is wrong, meaning that BBCNS does not correctly
Relay Node generate the DSs. Then, the solution soltmax is invalid at the
end of the iteration tmax . Assume we have the initial solutions
1: Input: round number (r ), parent list (P Li st).
sol1 = (d11 , . . . , dk1i ), sol2 = (d12 , . . . , dk2 ), …, sol p =
2: Output: parent id (P I D) that works as relay node.
(d1n , . . . , dkn ). According to the BBCNS, a group of solutions
3: if P Li st = ∅ then
are generated in the initialization process using GSP. During
4: p = |P List 1
| the fitness evaluation step, each of these solutions is evaluated
5: else
using the Equation 1, and the best solution with the highest fit-
6: p = 1
7: end if ness value is kept and saved for the subsequent operations. The
top best m solutions are then selected during the neighborhood
8: Set T empid = 0
search process and are subjected to swap mutation to generate
9: for every node si in P Li st & si .E > 0 do
new solutions. The newly generated solutions are evaluated
10: Set Tr = rand() {Random value}
based on their fitness, as discussed above. The remaining
11: if ω(si , r ) == 0 then
solutions are subjected to crossover operation in the global
12: calculate  using equation 3
search step to generate new solutions, which are evaluated and
13: Calculate T =  × p × (1 − p × mod(r,  p−1 ))−1
added to the population. These operations are repeated until
14: if Tr < T then
15: T empid = si .i d the maximum number of iterations tmax is reached. The best
solution with maximum fitness value and valid sets is retained
16: end if
in each iteration until the maximum number of iterations tmax
17: else
is reached. Moreover, the convergence of the Bees algorithm
18: ε(si , r ) = ε(si , r ) + 1 {consecutive rounds in which
has been proved in a number of studies [12], [47]. Therefore
si has not worked as relay node.}
BBCNS is convergent.
19: end if
Lemma 3: The formed tree by the second phase is con-
20: end for
nected.
21: if T empid = 0 then
22: Set P I D = T empid Proof: We present the proof through contradiction. Sup-
pose the formed tree is disconnected. During the second phase,
23: ω(si , r ) = r ound(1/ p) − 1
BS sends a message to neighbors specifying its hop distance
24: ε(si , r ) = 0
h B S . Each node si that receives the message updates its hop
25: end if
distance h si to h B S + 1, adds the sender i d to its parent list,
26: return P I D
and sends a message to its neighbors stating its hop distance
(h si ). Each neighbor node s j of si that receives this message
will execute the same steps executed by si , if (h si < h s j ).
D. The Correctness and the Evaluation of IDCT At the end of this process, each node will have a hop distance
In this section, we prove the correctness of our proposed and a list of parent nodes, implying that each node has a path
algorithm and evaluate its time and communication complex- to the BS, which contradicts the previous assumption.
ity. The proof and the evaluation are presented in a set of Lemma 4: The proposed IDCT prolongs the network life-
Lemmas as follows: time.
Lemma 1: The Generating Solution Procedure (GSP) gen- Proof : Network lifetime can be expressed as the sum of
erates a solution. the lifetimes of the generated disjoint DSs. The node with
Proof: We prove this by contradiction. Suppose GSP does minimum energy determines the lifetime of any DS, i.e., the
not generate any solution. In the proposed work, GSP is used network lifetime can be extended by considering disjoint sets
in the first phase (BBCNS) for initialization process. Since with maximum lifetime and this is the main idea of IDCT,
we assumed that GSP doesn’t generate a solution, there exist where IDCT’s fitness function is developed to form DS that
no valid DS in its solution. Meanwhile, according to GSP, all have nodes with convergence in their energy levels and thus
nodes are candidate nodes in the first iteration (say X is the maximizes the lifetime of the DS.
set of candidate nodes). GSP works by selecting a node s1 Lemma 5: If n is the number of nodes in the network
randomly from the set X, excludes it from X and adds it to a and l is the average number of children, the communication
set called D1 . The neighbors of the node s1 are then eliminated complexity of the IDCT will be O(n logl (n)) messages.
from X and added to a list called xD1. Proof : The number of messages exchanged during Col-
These operations will be repeated until X becomes empty. lector nodes selection (BBCNS), and Data gathering path
At the end of the iterations, we have two sets, the set of formation and collection phases can be computed as follows:
21106 IEEE SENSORS JOURNAL, VOL. 21, NO. 18, SEPTEMBER 15, 2021

• Collector Nodes Selection Phase: In this phase, every TABLE I

node sends its information to the BS. Thus, the number S IMULATION S ETTINGS
of messages exchanged in this phase will be O(n).
• Data Gathering Path Formation and Collection Phase: In
the initial step, a number of O(n) messages is exchanged
where each node sends a message containing its hop
distance to its neighbors. There will be no exchanged
messages during the parent node selection step, because
the selection is done locally by the node. At the same
time, O(n logl (n)) messages will be exchanged in the
data collection step, where the collector node constructs
a path to the BS by determining the parent node and the
selected parent node continues the path construction by 2.0 ≤ p ≤ 4.0 is the path loss factor and propagation loss
forwarding data to its next node towards the BS. is inversely proportional to d 2 or d 4 for shorter or longer
Therefore, in the worst case, a total of O(n logl (n)) messages distances respectively.
will be exchanged by our proposed IDCT. The parameters of Bees algorithm are as follows: Count of
Lemma 6: If n is the number of nodes in the network, l is the Selected Sites = 20, Count of Scout Bees = 40, Count of
average number of children, p is the population size, q is the selected Elite sites = 8, Count of recruited Bees for selected
average number of parent nodes, and tmax is the maximum sites = 8, maximum iterations = 200, and Count of recruited
number of iterations, then the total time complexity of the Bees for Elite sites = 16.
IDCT will be O(n 2 + pntmax logl ( p)) Furthermore, we assume that a count of 100 nodes are
Proof : The time complexity for the two phases (collector randomly distributed in a 100 × 100 m 2 area, with BS in the
nodes selection, and data gathering path formation and collec- corner. The simulation settings provided in Table I are used
tion) can be computed as follows [38], [39]: for the experiments.
• Collector Nodes Selection Phase: This phase is initiated Used Performance Metrics:
and executed by BS. The BS collects the nodes informa- 1) Network lifetime: the round at which the first (FND),
tion, then finds energy-aware disjoint DSs based on the half (HND), and the last node dies (LND), respectively.
Bees algorithm. The first step will take O(n 2 ) time to 2) Average consumption of energy per round: this is the
initialize the population using GSP. Then BS will take average energy consumed until the FND.
O( pntmax log p) time to perform neighborhood search 3) Standard deviation of energy: this metric gives the
and global search. standard deviation of expended energy until FND.
• Data Gathering Path Formation and Collection Phase: 4) Number of live nodes: the number of alive nodes per
The initialization process will take O(n) time, q × n round.
time for parent node selection, and q logl (n) time for data
collection, where the collector node constructs a path to
A. IDCT Evaluation
the BS by determining the parent node. Then the selected
parent node continues the path construction by forwarding In this section, we provide the impact of BS location,
the data to its next node towards the BS. communication radius, and node density on our proposed
IDCT. Then, the output of the proposed IDCT is compared
Therefore, the total time complexity of our proposed IDCT
to that of existing methods.
will be O(n 2 + pntmax logl p).
1) Impact of BS Location: The aim of this test is to demon-
strate the impact of the location of BS on IDCT performance.
IV. S IMULATION R ESULTS AND A NALYSIS
Here, in a 100 × 100 m area, 100 sensor nodes are
The performance evaluation of our proposed approach is deployed randomly. BS is placed at three different locations:
conducted by simulating in MATLAB R2016a. For confi- the corner (the coordinates (1, 1)), the middle of one edge
dence, we considered averaging the results using 40 different (the coordinates (1, 50)), and the center of the network (the
randomly generated WSNs by applying the MAX-DPA algo- coordinates (50, 50)). The performance is analyzed by using
rithm [13]. the same settings that was previously mentioned.
For the following tests, we consider the energy model in [2], The network lifetime is a critical concern in WSN and
[3], [25], [26], which is the widely accepted model. In this various definitions of network lifetime have also been used.
model, we can express the energy used to send and receive a The lifetime of a sensor network is most commonly defined
packet as follows: as FND, HND or LND. The time interval from the start of net-
E T x (s, d) = E elec × s + υ × s × d p . (5) work operation until the death of the first sensor node (FND)
is referred to as stability period of operation, and it is crucial
E Rx (s, d) = E elec × s. (6)
for many applications where the feedback from the network
Here, E elec is the electronics energy, υ is the expended must be reliable. Figure 5 depicts the effect of BS location
amplification energy to overcome multi-path/ free space loss, on IDCT network lifetime (given in terms of FND, HND and
s is the packet size, and d is the transmission distance. LND). In this figure, the network lifetime becomes longer as
OSAMY et al.: IDCT: INTELLIGENT DATA COLLECTION TECHNIQUE 21107

Fig. 5. Impact of BS location and communication radius on lifetime (FND, HND, and LND).

Fig. 6. Impact of BS location and communication radius on energy.

Fig. 7. Impact of BS location and communication radius on dead nodes per round.

Fig. 8. Impact of BS location and communication radius on standard deviation of energy per round.

the BS location is changed from corner to the middle of one of the BS in the center provides the longest lifetime. This is
edge, and then to the center of the area regardless of the change because the distance from the nodes to the BS decreases as
in communication radius. The results reveal that the location the BS moves from the corner to the center.
21108 IEEE SENSORS JOURNAL, VOL. 21, NO. 18, SEPTEMBER 15, 2021

Fig. 9. Impact of standard deviation, average energy and number of live Fig. 10. Impact of node density against average consumed energy and
standard deviation until first node dies and network lifetime.
nodes against number of rounds and node density.

As the main concern of WSN is the energy expended along the distribution of dead nodes and the standard deviation of
the network lifetime, we present the following results to show consumed energy during round operations. It is clear from
the IDCT performance in balancing energy utilization of the these figures that the average expended energy as well as the
nodes with change in communication radius and BS location standard deviation of energy consumption decrease as the BS
along its lifetime. Figure 6 provides the effect of BS location moves from the corner to the center. The reason for this is same
on the average energy expended until FND and the standard as discussed above, i.e. due to the impact of the changes in the
deviation of energy. It is clear that expended energy until FND number of neighboring nodes. It is evident from the figures that
and the standard deviation of consumed energy is reduced IDCT performs well on balancing the energy consumption
as the BS location is changed to the area center. Moreover, among nodes with the appropriate BS positioning, which in
the BS located at the corners and edges may have only a few turn ensures increased lifetime.
neighboring nodes compared to the center, resulting in less 2) Impact of Communication Radius: Our aim in this test
balance between the nodes, affecting the IDCT and shortening is to see how IDCT performs over various communication
the lifetime. radius (C R ) of nodes. 100 nodes are deployed randomly on a
To further analyze the network lifetime performance and 100 m × 100 m area. We change the C R from 25 to 45 with
stability period of operation, we conducted additional exper- an increment of 10 m with BS at the corner.
iments by changing the BS location and radius of com- The impact of C R on the lifetime of the network (in terms of
munication. Accordingly, the flow of dead nodes and the FND, HND and LND), the average energy consumed until the
standard deviation of consumed energy by nodes during the first node dies (FND), and the standard deviation of expended
rounds of operation are shown in Figures 7 and 8, respectively. energy respectively are already depicted in Figures 5 and 6.
The results indicate the impact of changing BS location on From these figures it is evident that as the C R increases,
OSAMY et al.: IDCT: INTELLIGENT DATA COLLECTION TECHNIQUE 21109

the standard deviation of energy consumed also increases,

at the same time, the WSN lifetime and average of the energy
expended until FND are decreased. It is also clear that the
network lifetime behavior does not change as the BS’s position
changes with different C R . It is because, the increase in C R
causes the increase in the neighborhood nodes’ count, resulting
in an increase in the formed DSs. Moreover, Figures 7 and 8
indicate the performance in terms of the distribution of dead
nodes and the standard deviation of consumed energy during
round operations. As the transmission radius grows, the nodes
count that could be covered by the DS also increases. As a
result, fewer nodes are needed to be added to the set. While the
transmission radius decreased, more nodes are added to the set
which leads to fewer normal nodes and more collector nodes.
Further, for higher transmission radius values, fewer collectors
are included in the set and each collector node has several
normal nodes. As the sensor transmission radius increased,
it is noticed that the WSN lifetime slightly decreases, within
3-5%. There is only a slight increase in energy consumption
per round.
3) Impact of Node Density: The impact of node density on
IDCT performance is examined in this test. Here we consider
100, 150, 200, and 250 nodes deployed in our simulation
region of size 100 × 100m randomly; with BS at the corner
and with the same settings mentioned previously.
The number of live nodes, the standard deviation of energy
expended by the nodes, and the average of energy consumed
by the nodes in each round of operation are presented in
Figures 9c, 9a and 9b, respectively for different node densities.
In addition to this, we analyzed the algorithm performance
in terms of energy until FND for different node densi-
ties. Figures 10a and Figure 10b depict the average energy
expended and the standard deviation of expended energy until Fig. 11. Network lifetime in IDCT, CDS-LEACH, LEACH-DS-ACO,
FND respectively. Moreover, the Figure 10c shows the impact FUZZY-DS-ACO,BEE-VBC, and CDS-FOR algorithms.
of node density on the lifetime of the network (in terms
of FND, HND and LND respectively). It is evident from
these figures that, as the node density increases, the energy
consumed increases while the WSN lifetime and the standard
deviation of expended energy decreases. The reason is that,
as the density increases, acquisition of data is also increased,
which leads to an increase in energy consumption.
4) Performance Evaluation Against Existing Methods: In this
section, we compare the proposed IDCT’s performance to
that of existing algorithms: CDS-LEACH, FUZZY-DS-ACO,
Fig. 12. Average expended energy until FND in IDCT, CDS-LEACH,
LEACH-DS-ACO, BEE-VBC and CDS-FOR using the above LEACH-DS-ACO, FUZZY-DS-ACO,BEE-VBC, and CDS-FOR.
performance metrics and settings.
Figure 11 shows the lifetime of the network (FND, HND DS-ACO, BEE-VBC and CDS-FOR up to 47%, 6%, 22%,
and LND respectively). From this Figure, it is evident that 31%,and 32%, respectively.
IDCT improves the lifetime of WSN in terms of the FND Figure 12 shows the average expended energy per round
compared with CDS-LEACH, BEE-VBC, LEACH-DS-ACO, until FND. This is computed by dividing total expended
FUZZY-DS-ACO and CDS-FOR up to 61%, 30%, 23%, energy until FND by the respective round number. From
24%, and 50%, respectively and shows improvement in life- Figure 12, we can see that IDCT evenly distributes the work-
time under HND compared with CDS-LEACH, LEACH-DS- load among the WSN nodes during parent selection and split
ACO, FUZZY-DS-ACO, BEE-VBC and CDS-FOR up to the workload by round. IDCT achieves a gain in reducing
51%, 13%, 31%, 27%, and 37%, respectively. Also, IDCT the consumed energy compared with CDS-LEACH, FUZZY-
improves the lifetime of the network in terms of the LND DS-ACO, LEACH-DS-ACO, BEE-VBC and CDS-FOR up to
compared with CDS-LEACH, FUZZY-DS-ACO, LEACH- 82%, 10%, 4%, 18%, and 34%, respectively.
21110 IEEE SENSORS JOURNAL, VOL. 21, NO. 18, SEPTEMBER 15, 2021

each round. In order to transmit collector data, an aggregation

tree is established and at each round, best relay is selected in
a dynamic way to achieve balancing, and extends the network
lifetime. The proposed IDCT technique is then compared with
other baseline methods and we obtained acceptable results.
The analysis revealed that the IDCT scheme outperforms the
existing baseline methods in terms of enhancing the WSN
lifetime.
ACKNOWLEDGMENT
Fig. 13. Standard deviation of energy per round in IDCT, CDS-LEACH, The authors gratefully acknowledge Qassim Univer-
LEACH-DS-ACO, FUZZY-DS-ACO,BEE-VBC, and CDS-FOR. sity, represented by the Deanship of Scientific Research,
on the financial support for this research under the
Figure 13 demonstrates the standard deviation of the energy number (ucc-bs-2020-1-3-I-10127) during the academic
expended by the nodes for the evaluated algorithms. The year 1441 AH /2020 AD.
results show that there exist a uniform distribution of energy
expended among the nodes in IDCT. This is because of the R EFERENCES
better balance in the WSN energy consumption provided by [1] M. Wollschlaeger, T. Sauter, and J. Jasperneite, “The future of industrial
IDCT compared to the other algorithms. IDCT gives the communication: Automation networks in the era of the Internet of
Things and industry 4.0,” IEEE Ind. Electron. Mag., vol. 11, no. 1,
lowest standard deviation followed by LEACH-DS-ACO (by pp. 17–27, Mar. 2017, doi: 10.1109/MIE.2017.2649104.
1%increased compared to IDCT) then FUZZY-DS-ACO (by [2] A. Aziz, K. Singh, W. Osamy, and A. M. Khedr, “An efficient compres-
2%increased compared to IDCT), while the others give an sive sensing routing scheme for Internet of Things based wireless sensor
networks,” Wireless Pers. Commun., vol. 114, no. 3, pp. 1905–1925,
increase by 4% compared to IDCT in standard deviation. This Oct. 2020, doi: 10.1007/s11277-020-07454-4.
indicate that IDCT, LEACH-DS-ACO, and FUZZY-DS-ACO [3] A. Aziz, K. Singh, W. Osamy, and A. M. Khedr, “Effective algorithm
balance the consumed energy per round. for optimizing compressive sensing in IoT and periodic monitoring
applications,” J. Netw. Comput. Appl., vol. 126, pp. 12–28, Jan. 2019,
In summary, CDS-LEACH gives the lowest network lifetime doi: 10.1016/j.jnca.2018.10.013.
performance since the DS is computed by recognizing the [4] W. Osamy, A. A. El-Sawy, and A. M. Khedr, “Effective TDMA
maximum degree of those nodes to discover the highest cover scheduling for tree-based data collection using genetic algorithm in
wireless sensor networks,” Peer-to-Peer Netw. Appl., vol. 13, no. 3,
nodes. FUZZY-DS-ACO and LEACH-DS-ACO follows the pp. 796–815, May 2020, doi: 10.1007/s12083-019-00818-z.
same concept (in which clusters are formed then using ACO [5] A. M. Khedr, A. Aziz, and W. Osamy, “Successors of PEGASIS pro-
the chain is established between nodes) with one difference tocol: A comprehensive survey,” Comput. Sci. Rev., vol. 39, Feb. 2021,
Art. no. 100368, doi: 10.1016/j.cosrev.2021.100368.
in the way of leader node selection, where in FUZZY-DS-ACO [6] C. Mahapatra, Z. Sheng, P. Kamalinejad, V. C. M. Leung, and
the selection is based on Fuzzy logic approach while LEACH- S. Mirabbasi, “Optimal power control in green wireless sensor net-
DS-ACO employs random selection approach. Dominated works with wireless energy harvesting, wake-up radio and trans-
mission control,” IEEE Access, vol. 5, pp. 501–518, 2017, doi:
sets in FUZZY-DS-ACO and LEACH-DS-ACO are formed 10.1109/ACCESS.2016.2644607.
per cluster which affects the performance of them. CDS- [7] Z. Zhang, J. Zhou, Y. Mo, and D.-Z. Du, “Performance-
FOR employs a k-connected DS to ensure that each node guaranteed approximation algorithm for fault-tolerant connected
dominating set in wireless networks,” in Proc. 35th Annu. IEEE
is connected to a minimum of k-nodes and gain reliability, Int. Conf. Comput. Commun. (INFOCOM), Apr. 2016, pp. 1–8,
which can be attributed to the corresponding increase in doi: 10.1109/INFOCOM.2016.7524456.
energy consumption per round. Compared with CDS-LEACH, [8] R. Wan, N. Xiong, and N. T. Loc, “An energy-efficient sleep scheduling
mechanism with similarity measure for wireless sensor networks,” Hum.-
FUZZY-DS-ACO, LEACH-DS-ACO, BEE-VBC and CDS- Centric Comput. Inf. Sci., vol. 8, no. 1, pp. 1–22, Jun. 2018, doi:
FOR, the proposed IDCT utilizes the benefit of Bees Based 10.1186/s13673-018-0141-x.
Collector Nodes Selection (BBCNS) to form disjoint sets that [9] T. Pino, S. Choudhury, and F. Al-Turjman, “Dominating set algo-
rithms for wireless sensor networks survivability,” IEEE Access, vol. 6,
work based on round-robin concept. Moreover, each collector pp. 17527–17532, 2018, doi: 10.1109/ACCESS.2018.2819083.
node selects its relay node considering to achieve a balance in [10] X. Kui, S. Zhang, J. Wang, and J. Cao, “An energy-balanced clustering
energy consumption, which improves the WSN lifetime than protocol based on dominating set for data gathering in wireless sen-
sor networks,” in Proc. IEEE Int. Conf. Commun. (ICC), Jun. 2012,
the other schemes. pp. 193–197, doi: 10.1109/ICC.2012.6363775.
[11] M. Bezoui, A. Bounceur, R. Euler, F. Lalem, and L. Abdelkader, “A new
V. C ONCLUSION algorithm for finding a dominating set in wireless sensor and IoT
The concept of sleep-awake scheduling by selecting disjoint networks based on the wait-before-starting concept,” in Proc. IEEE
DSs can extend the network lifetime by distributing data SENSORS, Oct. 2017, pp. 1–3, doi: 10.1109/ICSENS.2017.8233992.
[12] D. T. Pham, M. Castellani, and M. Sholedolu, “The bees algorithm
gathering tasks over these sets. And from this concept, a new and mechanical design optimisation,” in Proc. 5th Int. Conf. Infor-
solution named IDCT was introduced in this paper to enhance mat. Control, Automat. Robot. (ICINCO), 2008, pp. 250–255, doi:
the WSN lifetime performance. Here, we considered WSN 10.5220/0001506102500255.
[13] F. A. Onat and I. Stojmenovic, “Generating random graphs for
that have nodes with heterogeneous energy, and swarm-based wireless actuator networks,” in Proc. IEEE Int. Symp. World Wire-
Bees algorithm was adapted by introducing a procedure for less, Mobile Multimedia Netw., Espoo, Finland, Jun. 2007, pp. 1–12,
generating initial sets and implemented cross-over operation doi: 10.1109/WOWMOM.2007.4351712.
[14] S. Guha and S. Khuller, “Approximation algorithms for connected
during the global search step. The disjoint DSs that are formed dominating sets,” Algorithmica, vol. 20, no. 4, pp. 374–387, 1998, doi:
during the first phase (BBCNS) of IDCT work as collectors in 10.1007/PL00009201.
OSAMY et al.: IDCT: INTELLIGENT DATA COLLECTION TECHNIQUE 21111

[15] K. Islam, S. G. Akl, and H. Meijer, “Maximizing the lifetime [35] C. Patra, “Introducing connected dominating set as selection feature
of wireless sensor networks through domatic partition,” in Proc. of cluster heads in hierarchical protocols of wireless sensor networks,”
IEEE 34th Conf. Local Comput. Netw., Oct. 2009, pp. 436–442, doi: Global J. Comput. Sci. Technol., vol. 20, no. 1, pp. 21–26, 2020.
10.1109/LCN.2009.5355161. [36] S. Mondal, S. Ghosh, and U. Biswas, “A dominating set based
[16] I. Stojmenovic, M. Seddigh, and J. Zunic, “Dominating sets and neigh- data gathering in wireless sensor network using fuzzy logic and
bor elimination-based broadcasting algorithms in wireless networks,” ACO,” in Proc. Int. Conf. Informat. Anal., Aug. 2016, pp. 1–8, doi:
IEEE Trans. Parallel Distrib. Syst., vol. 13, no. 1, pp. 14–25, Jan. 2002, 10.1145/2980258.2980415.
doi: 10.1109/71.980024. [37] W. Shi, W. Liu, T. Wang, Z. Zeng, and G. Zhi, “Adding duty
[17] J. Wu and H. Li, “On calculating connected dominating set for efficient cycle only in connected dominating sets for energy efficient and fast
routing in ad hoc wireless networks,” in Proc. 3rd Int. Workshop data collection,” IEEE Access, vol. 7, pp. 120475–120499, 2019, doi:
Discrete Algorithms Methods Mobile Comput. Commun. (DIALM), 1999, 10.1109/ACCESS.2019.2937626.
pp. 7–14, doi: 10.1145/313239.313261. [38] A. Khedr and R. Bhatnagar, “Agents for integrating distributed data
[18] K. Islam, S. G. Akl, and H. Meijer, “A constant factor localized for complex computations,” Comput. Informat. J., vol. 26, no. 2,
algorithm for computing connected dominating sets in wireless sen- pp. 149–170, 2007.
sor networks,” in Proc. 14th IEEE Int. Conf. Parallel Distrib. Syst., [39] A. M. Khedr “Learning K-nearest neighbors classifier from distributed
Dec. 2008, pp. 559–566, doi: 10.1109/ICPADS.2008.78. data,” Comput. Inform., vol. 27, no. 3, pp. 355–376, 2008.
[19] J. P. Mohanty, C. Mandal, C. Reade, and A. Das, “Construction of [40] B. Abid, S. Messai, and H. Seba, “Energy-based connected dominating
minimum connected dominating set in wireless sensor networks using set for data aggregation for intelligent wireless sensor networks,” in
pseudo dominating set,” Ad Hoc Netw., vol. 42, pp. 61–73, May 2016, Proc. Int. Conf. Mach. Learn. Netw. Cham, Switzerland: Springer, 2018,
doi: 0.1016/j.adhoc.2016.02.003. pp. 193–211, doi: 10.1007/978-3-030-19945-6_13.
[41] V. S. Anitha and M. P. Sebastian, “A connected dominating set-
[20] B. Das, R. Sivakumar, and V. Bharghavan, “Routing in ad hoc networks
based weighted clustering algorithm for wireless sensor networks,” in
using a spine,” in Proc. 6th Int. Conf. Comput. Commun. Netw.,
Proc. IEEE Int. Conf. Wireless Commun., Netw. Inf. Secur., Jun. 2010,
Sep. 1997, pp. 34–39, doi: 10.1109/ICCCN.1997.623288.
pp. 530–534, doi: 10.1109/WCINS.2010.5544142.
[21] J. Yu, N. Wang, G. Wang, and D. Yu, “Connected dominating [42] H. Faheem, N. Ilyas, S. Ul, and S. Tanvir, “Connected dominating set
sets in wireless ad hoc and sensor networks—A comprehensive sur- based optimized routing protocol for wireless sensor networks,” Int.
vey,” Comput. Commun., vol. 36, no. 2, pp. 121–134, 2013, doi: J. Adv. Comput. Sci. Appl., vol. 7, no. 11, pp. 322–331, 2016, doi:
10.1016/j.comcom.2012.10.005. 10.14569/IJACSA.2016.071142.
[22] T. Shi, S. Cheng, Z. Cai, and J. Li, “Adaptive connected [43] J. Albath, M. Thakur, and S. Madria, “Energy constrained dominat-
dominating set discovering algorithm in energy-harvest sensor ing set for clustering in wireless sensor networks,” in Proc. 24th
networks,” in Proc. IEEE INFOCOM, Apr. 2016, pp. 1–9, IEEE Int. Conf. Adv. Inf. Netw. Appl., Apr. 2010, pp. 812–819, doi:
doi: 10.1109/INFOCOM.2016.7524504. 10.1109/AINA.2010.14.
[23] J. Yu, Q. Zhang, D. Yu, C. Chen, and G. Wang, “Domatic partition [44] X. Kui, J. Wang, S. Zhang, and J. Cao, “Energy balanced clustering data
in homogeneous wireless sensor networks,” J. Netw. Comput. Appl., collection based on dominating set in wireless sensor networks,” Adhoc
vol. 37, pp. 186–193, Jan. 2014, doi: 10.1016/j.jnca.2013.02.025. Sensor Wireless Netw J., vol. 24, pp. 99–217, Feb. 2015.
[24] O. Yilmaz and K. Erciyes, “Distributed weighted node shortest path [45] E. G. Julie and S. Tamilselvi, “CDS-fuzzy opportunistic routing protocol
routing for wireless sensor networks,” in Proc. Int. Conf. Wire- for wireless sensor networks,” Wireless Pers. Commun., vol. 90, no. 2,
less Mobile Netw. Berlin, Germany: Springer, 2010, pp. 304–314, pp. 903–922, 2016, doi: 10.1007/s11277-016-3250-8.
doi: 10.1007/978-3-642-14171-3_26. [46] G. V. S. Varsa and D. Sridharan, “A balanced energy efficient virtual
[25] W. Osamy, A. A. El-Sawy, and A. Salim, “CSOCA: Chicken backbone construction algorithm in wireless sensor networks,” AEU
swarm optimization based clustering algorithm for wireless sen- Int. J. Electron. Commun., vol. 107, pp. 110–124, Jul. 2019, doi:
sor networks,” IEEE Access, vol. 8, pp. 60676–60688, 2020, doi: 10.1016/j.aeue.2019.05.018.
10.1109/ACCESS.2020.2983483. [47] M. S. Bahari, “Enhancement of bees algorithm for global optimisation,”
[26] A. Salim, W. Osamy, A. M. Khedr, A. Aziz, and M. Abdel-Mageed, Ph.D. dissertation, Dept. Mech. Eng., School Eng. College Eng. Phys.
“A secure data gathering scheme based on properties of primes and Sci., Univ. Birmingham, Birmingham, U.K., 2019.
compressive sensing for IoT-based WSNs,” IEEE Sensors J., vol. 21, [48] D. T. Pham and M. Castellani, “A comparative study of the bees
no. 4, pp. 5553–5571, Feb. 2021, doi: 10.1109/JSEN.2020.3032585. algorithm as a tool for function optimisation,” Cogent Eng., vol. 2, no. 1,
[27] Y. Lim and S. Kang, “Intelligent approach for data collection in wireless Dec. 2015, Art. no. 1091540, doi: 10.1080/23311916.2015.1091540.
sensor networks,” Int. Arab J. Inf. Technol., vol. 10, no. 1, pp. 36–42, [49] W. Sun, M. Tang, L. Zhang, Z. Huo, and L. Shu, “A survey of using
2013. swarm intelligence algorithms in IoT,” Sensors, vol. 20, no. 5, pp. 14–20,
[28] P. V. P. Raj, A. M. Khedr, and Z. A. Aghbari, “Data gathering via 2020, doi: 10.3390/s20051420.
mobile sink in WSNs using game theory and enhanced ant colony [50] M. V. R. Sarobin and R. Ganesan, “Swarm intelligence in wireless
optimization,” Wireless Netw., vol. 26, no. 4, pp. 2983–2998, 2020, doi: sensor networks: A survey,” Int. J. Pure Appl. Math., vol. 101, no. 5,
10.1007/s11276-020-02254-x. pp. 773–807, 2015.
[51] R. V. Kulkarni, A. Forster, and G. K. Venayagamoorthy, “Compu-
[29] A. Thiruchelvi and N. Karthikeyan, “Pair-based sink relocation and route
tational intelligence in wireless sensor networks: A survey,” IEEE
adjustment in mobile sink WSN integrated IoT,” IET Commun., vol. 14,
Commun. Surveys Tuts., vol. 13, no. 1, pp. 68–96, 1st Quart., 2011,
no. 3, pp. 365–375, Feb. 2020, doi: 10.1049/iet-com.2019.0054.
doi: 10.1109/SURV.2011.040310.00002.
[30] G. S. Arumugam and T. Ponnuchamy, “EE-LEACH: Development
of energy-efficient LEACH protocol for data gathering in WSN,” Walid Osamy received the B.Sc. (Hons.), M.S.,
EURASIP J. Wireless Commun. Netw., vol. 76, pp. 1–9, Dec. 2015, doi: and Ph.D. degrees in computer science from the
10.1186/s13638-015-0306-5. Faculty of Science, Zagazig University, Egypt,
[31] R. Sridhar and N. Guruprasad, “Energy efficient chaotic whale opti- in June 2000, June 2006, and September 2010,
mization technique for data gathering in wireless sensor network,” respectively. From July 2001 to March 2009,
Int. J. Electr. Comput. Eng., vol. 10, no. 4, p. 4176, Aug. 2020, doi: he was involved in the Project of Network
10.11591/ijece.v10i4.pp4176-4188. Infrastructure and Management Information Sys-
[32] X. Yu, W. Xu, and C. Chen, “HDCDS-CDG: A hierarchically tems, Communication Information Technology
diffused connected dominating sets based compressed data gather- Center (CITC) and a member of Information
ing scheme,” J. Phys., Conf., vol. 1570, no. 1, pp. 12–56, 2020, and Communication Technology Projects (ICTP),
doi: 10.1088/1742-6596/1570/1/012056. Zagazig University. From December 2010 to Jan-
[33] X. Kui, J. Wang, and S. Zhang, “A data gathering algorithm based on uary 2019, he was an Assistant Professor with the Department of Com-
energy-balanced connected dominating sets in wireless sensor networks,” puter Science, College of Computers and Informatics, Benha University,
in Proc. IEEE Wireless Commun. Netw. Conf. (WCNC), Apr. 2013, Egypt. Since February 2019, he has been working as an Associate
pp. 1139–1144, doi: 10.1109/WCNC.2013.6554724. Professor with the Department of Computer Science, College of Com-
[34] S. Ghosh, S. Mondal, and U. Biswas, “A dominating set based puters and Artificial Intelligence, Benha University. Since March 2015,
modified LEACH using ant colony optimization for data gath- he has been working as an Assistant Professor with Qassim University,
ering in WSN,” in Proc. 2nd Int. Conf. Adv. Electr., Elec- Buraidah, Saudi Arabia. His research interests include computational
tron., Inf., Commun. Bio-Inform. (AEEICB), Feb. 2016, pp. 390–396, intelligence and in the field of the IoT [mobile computing and wireless
doi: 10.1109/AEEICB.2016.7538315. sensor networks (WSNs)].
21112 IEEE SENSORS JOURNAL, VOL. 21, NO. 18, SEPTEMBER 15, 2021

Ahmed Salim received the B.Sc. degree in com- Ahmed A. El-Sawy received the B.S. and M.S.
puter science and the M.Sc. degree in the area degrees from Cairo University in 2000 and 2003,
of distributed computing from Zagazig Univer- respectively, and the Ph.D. degree in computer
sity, Egypt, in June 2001 and December 2006, science from Suez Canal University in 2011.
respectively, and the Ph.D. degree in systems, From 2004 to 2011, he was a Lecturer with the
networks, and telecommunication devices from Faculty of Computers and Information Systems,
the Bonch-Bruvich University of Telecommunica- Taif University, Saudi Arabia. Since 2012, he has
tion, Saint-Petersburg, Russia, in October 2010. been a Lecturer and an Associate Professor of
From January 2012 to December 2012, he was Computer Science with Benha University since
a Consultant of the Information and Commu- 2017. Since 2017, he served as the Head of
nication Technology Project (ICTP), Zagazig Computer Science for the Faculty of Computers
University. From January 2011 to May 2019, he was an Assistant Pro- and Artificial Intelligence, Benha University, Egypt. His research inter-
fessor with the Department of Mathematics, Faculty of Science, Zagazig ests include artificial intelligence in computer network, bioinformatics,
University. Since June 2019, he was as an Associate Professor with chimoinformatics, deep learning, and optimization.
the Department of Mathematics, Faculty of Science, Zagazig University.
Since January 2014, he has been working as an Assistant Professor
with the Department of Computer Science, Faculty of Science and
Art, Qassim University, Al-mithnab, Saudi Arabia. His research interests
include decomposable algorithms, computing, and wireless networks.

Ahmed M. Khedr received the B.Sc. degree in

mathematics and the M.Sc. degree in the area
of optimal control from Zagazig University, Egypt,
in June 1989 and July 1995, respectively, and the
M.Sc. and Ph.D. degrees in computer science
and engineering from the University of Cincin-
nati, Ohio, USA, in July 1999 and March 2003,
respectively. From March 2003 to January 2004,
he was a Research Assistant Professor with
the ECECS Department, University of Cincinnati.
From January 2004 to May 2009, he was an
Assistant Professor with Zagazig University. From September 2009 to
September 2010, he was an Associate Professor with the Department of
Computer Science, College of Computers and Information Systems, Taif
University, Saudi Arabia. Since December 2014, he has been a Professor
with Zagazig University. From September 2010 to December 2019,
he was an Associate Professor and has been a Professor with the
Department of Computer Science, College of Computing and Informat-
ics, University of Sharjah, United Arab Emirates, since January 2020.
His research interests include wireless sensor networks, the Internet of
Things, and distributing computing. He was awarded the State Prize of
Distinction in Advanced Technology in June 2009, the Sharjah Islamic
Bank Prize of Distinction in Research in May 2013, and the University of
Sharjah Prize of Distinction in Research in April 2014.