Computers, Materials & Continua Tech Science Press

DOI:10.32604/cmc.2021.013303
Article

Intelligent Software-Defined Network for Cognitive Routing

Optimization Using Deep Extreme Learning
Machine Approach
Fahd Alhaidari1, Sultan H. Almotiri2 , Mohammed A.Al Ghamdi2 , Muhammad Adnan Khan3, * ,
Abdur Rehman4 , Sagheer Abbas4 , Khalid Masood Khan3 and Atta-ur-Rahman5

1
Department of Computer Information System, College of Computer Science and Information Technology,
Imam Abdulrahman Bin Faisal University, Dammam, 31441, Saudi Arabia
2
Department of Computer Science, Umm Al-Qura University, Makkah City, 715, Saudi Arabia
3
Department of Computer Science, Lahore Garrison University, Lahore, 54792, Pakistan
4
School of Computer Science, NCBA&E, Lahore, 54000, Pakistan
5
Department of Computer Science, College of Computer Science and Information Technology, Imam Abdulrahman
Bin Faisal University, Dammam, 31441, Saudi Arabia
*
Corresponding Author: Muhammad Adnan Khan. Email: [email protected]
Received: 30 July 2020; Accepted: 04 September 2020

Abstract: In recent years, the infrastructure, instruments, and resources of

network systems are becoming more complex and heterogeneous, with the
rapid development of current internet and mobile communication technolo-
gies. In order to efficaciously prepare, control, hold and optimize networking
systems, greater intelligence needs to be deployed. However, due to the inher-
ently dispensed characteristic of conventional networks, Machine Learning
(ML) techniques are hard to implement and deployed to govern and operate
networks. Software-Defined Networking (SDN) brings us new possibilities to
offer intelligence in the networks. SDN’s characteristics (e.g., logically cen-
tralized control, global network view, software-based site visitor analysis, and
dynamic updating of forwarding rules) make it simpler to apply machine learn-
ing strategies. Various perspectives of fiber-optic communications including
fiber nonlinearity coverage, optical performance checking, cognitive short-
coming detection/anticipation, and arranging and improvement of software-
defined networks are examined in Machine Learning (ML) applications.
This research paper has presented an imaginative framework concept called
Intelligent Software Defined Network (ISDN) for Cognitive Routing Opti-
mization (CRO) using Deep Extreme Learning Machine (DELM) approach
(ISDN-CRO-DELM) in light of the new challenges in the development and
operation of communication systems, and capturing motivation from how
living creatures deal with difficulty and usability. The proposed methodology
develops around the planned applications of progressive DELM methods
and, specifically, probabilistic generative models for framework wide learn-
ing, demonstrating, improvement, and information description. Furthermore,

This work is licensed under a Creative Commons Attribution 4.0 International License,
which permits unrestricted use, distribution, and reproduction in any medium, provided
the original work is properly cited.
1270 CMC, 2021, vol.67, no.1

ISDN-CRO-DELM, suggest to integrate this learning framework with the

ISDN for CRO and reconfiguration approaches at the system level. MATLAB
2019a is used for DELM simulation and superior results show the effectiveness
of the proposed framework.

Keywords: SDN; DELM; machine learning; cognition

1 Introduction
In recent times, data processing is growing exponentially in the world, with the accelerated
growth of mobile devices (e.g., cell phones, mobile vehicles and smart home appliances) as well
as network technology (e.g., cloud storage and edge computing) [1]. Networks are becoming
heterogeneous and it is a challenge to automate the flow of traffic and control a large number
of devices [2]. In general, the production network uses many devices, operates a number of
procedures and supports a number of applications [3,4]. As an example, different types of modules
have been used in wireless networks, such as ZigBee, WiMAX, IEEE 802.11 ac/ad, Bluetooth,
and Long-Term Evolution (LTE), with specific communication range, strength, and operating
mechanisms and numerous communication technologies. Further, transparency in networks is one
way to solve these problems. A few years ago, the Knowledge Plan (KP) proposed an approach
to develop automation, advice and wisdom on the Internet through the use of ML and cognitive
technologies [5]. However, the KP is still not prototyped or implemented at the time of this
writing. The main reason is that conventional networks are fundamentally distributed, where each
node, such as a router or switch, can only access and operate on a small network ratio. It’s really
hard to learn from the networks that they have only a partial knowledge of the entire network
to manage the entire domain [6]. Luckily, recent developments in the Software Defined Network
(SDN) domain will enable learning more complex.
In the other direction, this will enable us to develop massive information and communication
technologies (ICTs) [7]. Even today, no one can imagine our life without the Internet [8]. It is even
hard to consider that the increasing accessibility of connected products means that today the
Internet has not gained its importance by enabling people to discover data and providing us with
innovative facilities in our daily lives [9]. According to the Internet of Things (IoT) European
Research Cluster, IoT is a vibrant, global network infrastructure that automatically configures
and communicates physical and virtual objects through standard and interoperable procedures.
IoT would have been the best hope for a viable city [10]. In order to make IoT smart, a
number of computational technologies use IoT. Considering a wide range of IoT applications,
data mining, artificial intelligence, cloud computing and neural networks are among the most
important technologies [11].
Cognitive networks focus on the “cognitive loop” as shown in Fig. 1. The Sense, Plan,
Decision, Act, follow a full cognitive protocol with learning and environmental input. The state of
sense gathers, tracks, and pre–processes information. The plan state then interprets the collected
data in the framework and uses the findings for scheduling purposes. The state of Decision
sets out the measures and analyzes information for possible options. Eventually, the active state
imposes the actual operation of the network. Environmental data is used to restart the loop and
in the final control, alarms and alerts can also be sent. All previous states provide feedback that
improves the efficiency and response of the learning algorithm under the specified policies.
Some of the work on cognitive networks suggested that cognitive entities may use the best
tools to perform optimization tasks [12,13], while others have used special reasoning techniques:
CMC, 2021, vol.67, no.1 1271

they have also not explained clearly why these techniques should be chosen [14]. Over the last
decade, numerous published papers have explored the implementation of various ML algorithms
with a view to improving the limited management capabilities of legacy SDNs.

Figure 1: The cognition loops

The Internet of Things (IoT) transforms communities around the world into “smart cities”
by creating a new way of living in urban areas [15]. Main advantages include improved safety,
health, improved environment for education and housing, energy consumption, better climate and
ecosystem efficiency, green economy and better employment [3]. Although the central knowledge
of smart cities has been around for more than a decade, the definition has improved a lot
since its initial release, it is now expected to fundamentally change city life as a number of
important facilitators. Hall et al. [16] claimed that the vision of smart cities for all systems, such
as electricity, water and logistics, is a clean, stable, green and efficient future metropolitan center.
Bakıcı et al. [17] indicated that smart city is an innovation-intensive and better city that combines
people, information and urban features with innovative technology to create a healthy and safe
city with competitiveness and innovative commerce to improve quality of life. Giffinger et al. [18]
argued that the smart city is an ingenious community in which a number of components, such as
individuals, ecosystems, flexibility, democracy and the economy, are integrated into a smart system.
In our research, SDN routing optimization is the primary concept for the implementation
of complicated network tracking systems or malicious behavior or policy infringement systems.
This work summarizes the problems of SDN routing and smart networking, and then defines a
coherent framework for optimizing the software-defined network. The DELM method will be used
to make software-defined networks more secure with improved performance. Safe and efficient
1272 CMC, 2021, vol.67, no.1

networks bring together potential smart cities with advanced systems and policies to adapt to
demands [19].
The control plane and the data plane shall be separated by the SDN. The network resources
in SDN are managed by the networking operating system, which serves as a logically centralized
controller. The network must be dynamically managed by the SDN controller. In addition, the
centralized controller has a global perspective on the system by monitoring and assembling the
status and arrangement of the system information as well as packet and stream sized data.
The implementation of SDN ML methods is acceptable and effective for the following rea-
sons. First, the recent development of computer technology such as the Graphical Processing Unit
(GPU) and the Tensor Processing Unit (TPU) provides a decent opportunity to use the exciting
ML networking area (e.g., deep neural networks) [20,21]. Second, data is essential for data-driven
machine learning algorithms. The concentrated SDN controller has a global network opinion
and is able to gather numerous network information that makes the implementations of ML
algorithms easier. Third, ML methods can transport SDN controller intelligence based on data
assessment, network optimization in addition to the automated delivery of network facilities based
on the historical network, and real-time information. Finally, SDN technology allows optimized
network solutions (e.g., configuration and resource allocation) to be used in real-time on the
network using machine-learning algorithms [22].
Present cognitive science theory has presented various approaches that conflict with the way
they tackle the relationship between sensory-motor and cognitive processes. This research adds
to the dialog by claiming that the investigation is appropriately expressed as a developmen-
tal investigation and that, eventually, in order to perceive the properties of human cognition,
the researcher asked how human cognition had these properties. The researcher assumes that,
since advancement creates brains, bodies, and conditions for cognition, it is inflexibly linked to
procedures for detecting and acting invisible to them.
This article examines the state-of-the-art ML strategies that can be developed and imple-
mented in SDN. The review will focus on the use of ML methods to enhance the efficiency,
intelligence, effectiveness and safety of SDN, followed by a concise diagram of things to inquire
about presented solutions in related regions. The proposed model is included in the methodology
as shown in Fig. 2. This paper examines the DELM for optimization of the software-defined
network in order to achieve optimum accuracy. A data set, NSL-KDD, with 47840 data samples
is used in the training and testing of SDN optimization with DELM, so that each sample has spe-
cific and diverse features. The analysis and comparison shall be carried out using state-of-the-art
approaches of machine learning.

2 Related Works
ML applications have attracted a lot of attention. Patcha et al. [23] have provided a thorough
report of the applications for intrusion detection ML methods. Nguyen et al. [24] have reported
a detailed description on the classification of IP traffic using ML. Bkassiny et al. [25] have
researched many complex learning difficulties in Cognitive Radio Networks (CRNs) and analyzed
current ML-based solutions. How ML approaches can be utilized to solve common problems in
the network of wireless sensors has been investigated in [26]. Wang et al. [27] presented state-of-
the-art techniques to develop heterogeneous networks in Artificial Intelligence (AI) and discussed
future research problems. Buczak et al. [28] investigated cybersecurity intrusion detection methods
for ML and Data Mining (DM). Klaine et al. [29] have investigated classification & comparison of
CMC, 2021, vol.67, no.1 1273

machine-learning systems and their results in cellular networks. Fadlullah et al. [30] have explored
how ML methods can be used to enhance network traffic organizations. Similar to [23], Hodo
et al. [31] have focused on the ML-based Intrusion Detection System (IDS). Zhou et al. [32] have
investigated on the utilization of ML and cognitive radiation innovation to develop the wireless
2 network spectrum and enhanced its energy efficiency. Chen et al. [33] have researched solutions
based on neural networks to address wireless network problems such as connectivity, virtual reality
and edge caching.
Tang et al. [34] have managed to incorporate artificial intelligence in fog computing to facil-
itate intelligent large-scale data discovery. Feng et al. [35] have presented a theoretical paradigm
for smart households utilizing theoretical analytics, IoT frameworks, Bayesian Inference and Rein-
forcement Learning (RL) models in their cognitive memory. Cognitive IoT (CIoT) is another work
initiative designed to build a cognitive framework for IoT applications [36]. Vlacheas et al. [37]
have proposed a cognitive management system that would enable intelligent objects to interact
and benefit end-users.
Moore et al. [38] have predicted that by 2050, 70% of the world would reside in cities. The
smart concept is used to build comfort in your life. Smart City is an initiative that uses new tech-
nologies to enhance the efficiency of public life. Smart cities also enhance environmental efficiency
and provide residents with optimum amenities [39]. Transmission and computer technology are
essential for the promotion and transformation of urban areas into smart cities [40]. The Internet
of Things (IoT) integrates a wide range of physical objects, vehicles, smartphones and devices
that can communicate and share data with equipment, apps, controllers, sensors and the network.
Data mining approaches and machine learning algorithms should be equipped to collect details
and expertise on the development of data waste [41].
SDN has received widespread attention in recent years. The Open Networking Foundation is a
non-profit consortium that aims to develop standardize SDNs. The Open Networking Foundation
(ONF) describes the SDN as follows, “The control plane, data plane and network intelligence
are logically decoupled, network intelligence and the underlying network architecture is logically
centralized and excluded from applications” [42].
With regard to routing optimization, Francois et al. [43] have designed Random Neural
Networks with Reinforcement Learning (RL) based centralized Cognitive Routing Engine (CRE)
that recognizes the optimized network paths. Raimundo-Neto et al. [44] have proposed a wireless
network with a cognitive optical wireless system of 2.4 and 5 GHz. Shan et al. [45] have suggested
a centralized wireless cognitive compression sensing algorithm. Li et al. [46] have proposed an
enhanced learning approach that is implemented in cognitive wireless through fiber to introduce
dynamic channel selection. All processing capabilities are based on Cognitive Access Point (CAP),
which gathers snapshots of different Routing Area Update (RAU).
Recent research [47,48] on the importance of cognitive use in femtocells (small cells) research
shows that Quality of Service (QoS) is compromised with potential interferences. It also stresses
opportunistic cellular network cooperation. Therefore, an effective network management system
is required for cognitive deployment. SDN is an active area of research that can support a high
efficiency and provide intelligent control of the network.
Although the performance of the SDN in various organizations has been encouraging, the
separation of the control plane from the data plane increases the likelihood that the SDN applica-
tion would be exposed to multiple challenges and risks [49]. As a result, SDNs are susceptible to
a number of network infringements, including volumetric threats. These attacks will bring security
1274 CMC, 2021, vol.67, no.1

concerns to the three layers of SDN. There are several security mechanisms to detect and mitigate
Denial of Service (DoS) attacks on the SDN network as presented in [50]. Such strategies are
constructed by collecting the flow characteristics of the packets received, such as the average
number of packets, the Internet Protocol (IP) address size, the average number of bytes, etc.
In case of the malicious traffic, the monitoring packages are larger in size than the threshold
or borderline of the decision as described above [51]. Although current techniques do not require
a high level of data flow management and monitoring tools, they cannot protect the network for a
long time. The main drawbacks of the current situation are the rapid increase in the information
and the large data packets to be managed in real time. Due to a similarity between good traffic
and malicious packets, it is difficult to detect malicious traffic. The intruder will easily adjust the
packet header to mimic the usual traffic and mislead the protection systems. As anomalies are
constantly evolving, the attacker can make simple changes that will make it possible to use this
technique with new attacks. A lot of research has been performed in a number of areas, but only
a few have used SDN [52].

3 Proposed Model
3.1 System Model
The concept describes a high-level SDN architecture consisting of three primary planes, the
information plane, control plane, and application plane. The structural modules and connections
of each plane have been presented in Fig. 2.

Figure 2: Proposed model for software defined network routing optimization

The data plane is the lowest layer in the SDN framework and is also referred to as the
infrastructure plane. This plane consists of transmission gadgets consisting of physical switches
and virtual switches. Virtual switches are built-in programming tools that can run within normal
CMC, 2021, vol.67, no.1 1275

operating frameworks. These switches are responsible for delivering, dropping, and modifying
packets based on the Control Plane (CP) policy. The control plane is connected to the Southbound
Interfaces (SBIs) by means of which the control unit can control the processing and transmission
capabilities of the data plane.
The control plane is the intelligence of SDN schemes that can program network resources,
dynamically change security regulations and make network management scalable and agile. The
central module of the control unit is a reasonably central control plane that manages communi-
cation between transmission units and applications. Next, the controller uncovered and outlined
information from the data plane to the application plane. On the other hand, the controller is
modifying the request into custom strategies and disseminating it to sending units. The controller
also offers vital features, along with shortest route routing, network topology stockpiling, gadget
design, and national data warnings, that a considerable number of system applications require.
Fig. 3 shows the Cognitive Software Decision (CSD) is comprised of four layers named sensing
control layer, data semantic layer, decision making layer, and service evaluation layer.

Figure 3: Proposed model for cognitive software decision

3.1.1 Sensing Control Layer

This is the fundamental layer that contains the significant participation of this research. This
layer is a cognitive software designed inspired layer that has the ability to detect the ground similar
to a human’s performance. The entire CSD is comprised of the sensing process.

3.1.2 Data Semantic Knowledge Layer

In view of the information that has been sent from the sensing control layer, this layer changes
the information into knowledge utilizing semantic reasoning and knowledge learning. The data is
sent to the decision-making layer.

3.1.3 Decision Making Layer

This layer is accountable for making the decision based on the gathered data from the data
semantic layer. This layer can establish itself in which movement is dependent on the present state.

3.1.4 Service Evaluation Layer

This layer is inclined to negotiating the service of the sensors and furthermore liable for
assessing its display based on the service provider.
1276 CMC, 2021, vol.67, no.1

The application plane consisting of business applications is the top layer within the SDN
framework. These applications can provide new facilities and optimize business processes. In gen-
eral, the NBIs of the controllers allow applications to obtain the necessary network information.
Applications may implement the control logic to modify the actions of the network, based on the
data received and the business needs.
Fig. 4 illustrates the number of hidden layers, hidden neurons, and different forms of acti-
vation functions that have been employed in the Deep Extreme Learning Machine (DELM)
to obtain the optimal architecture for enhancing the software-defined network. The suggested
methodology consists of three separate layers, specifically the data collection, pre-processing,
and the application layer. The application layer consists of two sub-layers in which one layer
is for prediction and the other layer is for evaluation. Real data was obtained from sensors
for experimental research. After the collection of data from the sensor, the data is delivered as
feedback to the data collection level. In the pre-processing layer, numerous information cleaning
and analysis processes were applied to eliminate irregularities in the data. In the application layer,
DELM has been utilized for software-defined network optimization.

Figure 4: Systemic illustration of a deep extreme learning machine

3.2 Deep Extreme Learning Machine

The Deep Extreme Learning Machine (DELM) can be used in various fields to predict health
problems, energy consumption forecasts, transport and traffic management, etc. [53–55]. DELM
CMC, 2021, vol.67, no.1 1277

can be widely used for identification and regressive objectives in various ways, as it learns quickly
and improves the level of framework complexity.
Extreme Learning Machine (ELM) is a feed-forward neural network, which means that input
only travels one direction across a set of layers, but we have also used the back-propagation
approach in this predictive framework through the training process, where data streams back
across the network and back-propagation, the neural network adjusts weights to achieve high
accuracy at minimal error. Weights are constant during the validation phase of the network in
which the qualified model is imported and actual data is predicted. Information layer, numerous
hidden, with one output layer in the DELM model. Fig. 4 shows the fundamental model of a
DELM in which m is described as input,  describes hidden, and op denotes output layer nodes.
Mean Square Error (MSE) measurement to optimize software-defined network illustrated
during the evaluation phase. In the deep extreme learning machine method, the input, hid-
den, and output node are notated as an , pi , and gm , correspondingly, while all the N input
nodes, l hidden nodes, and M output nodes can be illustrated as a = a1 , a2 , . . . , aN T €RN ,
p = p1 , p2 , . . . , pl T €Rl , g = g1 , g2 , . . . , gM T €RM , correspondingly. The DELM framework will
therefore be efficiently portrayed as:
p = f (Ba + c) (1)

And
g = qp (2)
where B = [bin ] €Rl×N , c = c1 , c2 , . . . , cl T €Rl , Q = [qmi ] €Rm×l , and the activation function f ( )
could be employed as sigmoid, linear Gaussian prototypes, etc.
Assume here are V different training data, and allow av €RN and gv €RM and denote the vth
training and the resulting vth training outcome, correspondingly, where v = 1, 2, . . . , V . In the
training set the input structure and outcome structure presented as:
A = a1 , a2 , . . . , aV T €RN×V (3)

And
G = g1 , g2 , . . . , gV T €RM×V (4)

Correspondingly. We are able to replace Eq. (3) into Eq. (1) to acquire
 
P = f BA + 1T ⊗ c (5)

where P = p1 , p2 , . . . , pV T €Rl×V is the assessment categorization of all l hidden nodes, and ⊗
Kronecker product Rewrite Eqs. (4) and (5) in Eq. (2) to attain the real training sequence.
G = QP (6)

In DELM, only output weight Q is adaptable, though B (for example the input weights) and
c (for example the prejudices of the hidden nodes) are randomly focused. Label the projected
outcome as Y . Afterward DELM purely reduces the evaluation error.
E = Y − G = Y − QP (7)
1278 CMC, 2021, vol.67, no.1

By considering a minimum combination of squares Q for the situation

min min
E2F = Y − QP2F (8)
Q Q
where ·F indicates the Frobenius norm. For the problematic Eq. (8), an outstanding least-square
minimum standard solution is:
 −1
Q = YPT PPT (9)

To prevent overfitting, it is possible to use the common Tikhonov regularization to adjust

Eq. (9) into
 −1
Q = YPT PPT + v20 I (10)

where v20 > 0 indicate the regularization phrase. The backpropagation method involves the con-
figuration of weight, forward propagation, backward propagation of error, and the updating of
distinguishability. An activation function like g (x) = sigmoid is present in the hidden layer on each
neuron. This helps to design the sigmoid input function and the DELM hidden layer.
1  
E= sj − wpj 2 (11)
2
j

Sj = Desired output
wpj = Computed output

Eq. (11) specifies a backpropagation error, which can be measured by splitting the square
summation from the required outcome by 2. The adjustment in weight is needed to mitigate the
common error. The weight shift rates are shown for the output layer in Eq. (12).
∂R
Hi,j
l=6
∝− (12)
∂Hl=6
where i = 1, 2, . . . , 10 (Neurons) and j = outcome Layer
∂R
Hi,j
l=6
= −const (13)
∂Hl=6
Write Eq. (13) through the chain rule technique:

∂R ∂wplj ∂NhHjl
Hi,j
l=6
= −const × × (14)
∂wplj ∂NhHjl ∂Hi,j
l

The change weight value can be obtained by replacing the values in Eq. (13) as indicated
in Eq. (14).
     
Hi,j
l=6
= const sj − wpj × wplj 1 − wplj × wplj (15a)
CMC, 2021, vol.67, no.1 1279

From wp to H6
Hi,j
l=6
= const j wplj (15b)

The calculation for the related weight shift to the hidden weight is shown in the following
segment. It is quite complicated, because it may contribute to errors per weighted association on
the growing node.
From H6 to H1 or Hn
where n = 5, 4, 3, 2, 1
⎡ ⎤
 ∂R ∂wpl ∂Hl
∂wpln ∂NhHln
Hli,n ∝ − ⎣ ⎦×
j j
× × × (16a)
j
∂wplj ∂NhHlj ∂wpln ∂NhHhln ∂Hli,n
⎡ ⎤
  
Hli,n = R ⎣ l ⎦
j (Hn,j ) × wpln 1 − wplj × Zi,n (16b)
j

Hli,n = R n Zi,n (16c)

where
⎡ ⎤
  
n = ⎣ l ⎦
j (Hn,j ) × wpln 1 − wpln (16d)
j

The method for increasing the weight and the bias among the performance and the hidden
layer is illustrated in Eq. (16e).
l=6
Hi,j (t) = Hi,j
l=6
(t) + λHi,j
l=6
(16e)

Eq. (17) Demonstrates how weight and difference amongst input and hidden layer can
be changed.
l
Hi,n (t) = Hi,n
l
(t + 1) + λHi,j
l
(17)

4 Simulation and Results

The DELM method was employed in the suggested article [56] using MATLAB 2019. The
proposed solution based on DELM was used to train and fit the 148516 information instances.
This information is distributed randomly to 85% of the training (125973 records), and 15% of
the records are used for validation (22543 records). The data is pre-processed in order to remove
anomalies and error values. Along these lines, the proposed system has used a large number of
neurons and a variety of dynamic capacities are used in hidden layers. The proposed DELM is
implemented in order to investigate the viability of the proposed framework. In order to evaluate
the DELM approach, we have used the accuracy and misrate of the factual estimates as described
in the following equations (Eqs. (18), (19)).
O1 O0
Misrate = + (18)
T0 T1
1280 CMC, 2021, vol.67, no.1

 
O0
T0 +O
T
1
1
Accuracy = (19)
T0 + T1

In Eqs. (18) and (19), O signifies the projecting outcome of SDN and T denotes the real
outcome. O0 and T0 show that the situation is normal and no attack is found in the predictive
output and actual output respectively. O1 and T1 represents the attack is found in the predictive
O0 O1
and actual outputs respectively. and signifies that predictive and real output are same.
T0 T1
O1 O1
Similarly, and signifies the difference between the predictive and real outputs.
T0 T0

Table 1: Training of the suggested DELM based software defined network framework during the
prediction of intrusion
Proposed DELM based SDN
(85% of Data)
Total No. of Samples (N = 125973) Result (Output) (O0 , O1 )
Input Expected output (T0 , T1 ) O0 (Normal) O1 (Attack)
T0 = 67343 Normal 65254 2089
T1 = 58630 Attack 2673 55957

Tab. 1 shows the proposed DELM based software-defined network model that predicts intru-
sion detection during the training phase. In training, a total of 125973 samples are included,
which are then divided into 67343 and 58630 for normal and attack conditions. It is observed
that 65254 normal-class samples without any attack are correctly predicted and 2089 records are
misclassified. Likewise, a total number of 58630 samples are obtained in the event of an attack,
in which 55957 samples are correctly predicted and 2673 records are found to be incorrect.
Tab. 2 shows the proposed DELM based software-defined network model that predicts intru-
sion detection during the validation stage. In validation, a total of 22543 samples are included
that are split into 9710 and 34154 for normal and attack conditions. It is observed that 9194
samples of the normal class are correctly predicted and 516 records are misclassified. Similarly,
out of total 34154 samples, 33031 samples are correctly predicted and 1123 records are found to
be incorrect.
Tab. 3 demonstrates the proposed DELM-based software-defined network architecture output
in terms of accuracy and error rate across the training and validation process. It is shown that the
projected DELM based software-defined network system in training achieves 96.22% and 3.78%
accuracy and miss rate correspondingly. In across validation, the proposed DELM based software-
defined network system attains 92.73% and 7.27% accuracy and miss rate, respectively.
Tab. 4 shows the performance of the proposed DELM based software defined network model
using DELM in comparison to previous approaches in the literature [57,58].
Tab. 4 shows the comparison of attacks in SDN for the proposed DELM model with previous
methods as proposed by Li et al. [57] and Cui et al. [58]. Accuracy and Miss Rate metrics are
CMC, 2021, vol.67, no.1 1281

used to measure the evaluation of the proposed DELM based software-defined network system
model with l [57,58].
The proposed DELM based Software Defined Network model achieves the best classification
results of 96.22% and 92.73% for accuracy whereas 3.78% and 7.27% miss rate are computed in
training and validation phases respectively.

Table 2: Validation of the suggested DELM based software defined network framework during
the prediction of SDN attack
Proposed DELM based SDN
(15% Data)
Total No of Samples (N = 22543) Result (Output) (O0 , O1 )
Input Expected Output (T0 , T1 ) O0 (Normal) O1 (Attack)
T0 = 9710 Normal 9194 516
T1 = 34154 Attack 1123 33031

Table 3: Performance assessment of proposed deep extreme learning machine-based software

defined network system model in validation & training
Accuracy Miss rate
Training 96.22% 3.78%
Validation 92.73% 7.27%

Table 4: Comparison results of the DELM based software defined network framework
with literature
Literature Training Validation
Accuracy (%) Miss rate (%) Accuracy (%) Miss rate (%)
Li et al. [57] 84.81 15.19 83.2 16.8
Cui et al. [58] 93 7 87 13
Proposed DELM based 96.22 3.78 92.73 7.27
SDN system model

5 Conclusion
In this research, the proposed methodology has introduced an innovative concept called
Intelligent Software Defined Networking (ISDN) for Cognitive Routing Optimization (CRO) using
Deep Extreme Learning Machine (DELM) approach (ISDN-CRO-DELM). The proposed research
has employed DELM technique for simulation and attained better results in comparison to the
previous techniques. Considering past approaches, the proposed DELM technique outperforms
them and achieves superior performance. The proposed technique exhibits 96.22% accuracy that
is far better than the existing techniques. In future, our proposed technique will be expanded
1282 CMC, 2021, vol.67, no.1

by including more datasets and new machine learning techniques, structures, and algorithms will
be employed.

Acknowledgement: Thanks to our families & colleagues who supported us morally.

Funding Statement: This Work is supported by Data and Artificial Intelligence Scientific Chair at
Umm AlQura University.

Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding
the present study.

References
[1] A. Atta, S. Abbas, M. A. Khan, G. Ahmed and U. Farooq, “An adaptive approach: Smart traffic
congestion control system,” Journal of King Saud University-Computer & Information Sciences, vol. 32,
no. 9, pp. 1012–1019, 2020.
[2] H. Chourabi, T. Nam, S. Walker, J. R. Gil-Garcia, S. Mellouli et al., “Understanding smart cities:
An integrative framework,” in Proc. of 45th Hawaii Int. Conf. on System Science, Maui, HI, pp. 2289–
2297, 2012.
[3] A. Fatima, M. A. Khan, S. Abbas, M. Waqas, L. Anum et al., “Evaluation of planet factors of smart
city through multi-layer fuzzy logic (MFL),” ISC International Journal of Information Security, vol. 11,
no. 3, pp. 51–58, 2019.
[4] M. Naphade, G. Banavar, C. Harrison, J. Paraszczak and R. Morris, “Smarter cities and their
innovation challenges,” Computer, vol. 44, no. 6, pp. 32–39, 2011.
[5] D. D. Clark, C. Partridge, J. C. Ramming and J. T. Wroclawski, “A knowledge plane for the inter-
net,” in Proc. of the 2003 Conf. on Applications, Technologies, Architectures, & Protocols for Computer
Communications, Germany, pp. 3–10, 2003.
[6] A. Mestres, A. Rodriguez-Natal, J. Carner, P. Barlet-Ros, E. Alarcon et al., “Knowledge-defined
networking,” ACM SIGCOMM Computer Communication Review, vol. 47, no. 33, pp. 2–10, 2017.
[7] M. A. Khan, M. Umair and M. A. S. Choudhry, “GA based adaptive receiver for MC-CDMA sys-
tem,” Turkish Journal of Electrical Engineering & Computer Sciences, vol. 23, no. 1, pp. 2267–2277, 2015.
[8] S. Abbas, M. S. Khan, K. Ahmed, M. Abdullah and U. Farooq, “Bio-inspired neuro-fuzzy based
dynamic route selection to avoid traffic congestion,” International Journal of Scientific & Engineering
Research, vol. 2, no. 6, pp. 284–289, 2011.
[9] N. Naz, S. Abbas, M. A. Adnan, B. Abid, N. Tariq et al., “Efficient load balancing in cloud computing
using multi-layered mamdani fuzzy inference expert system,” International Journal of Advanced Computer
Science & Applications, vol. 10, no. 3, pp. 569–577, 2019.
[10] M. Dohler, I. Vilajosana, X. Vilajosana and J. Llosa, “Smart cities: An action plan,” in Proc. of
Barcelona Smart Cities Congress, Barcelona, Spain, pp. 1–6, 2011.
[11] R. Khare and P. Shrivasta, “Data mining for the Internet of Things,” in Exploring the Convergence of
Big Data & the Internet of Things, Chicago: IGI Global, pp. 181–191, 2018.
[12] R. Thomas, D. Friend, L. Dasilva and A. Mackenzie, “Cognitive networks: Adaptation and learning
to achieve end-to-end performance objectives,” IEEE Communications Magazine, vol. 44, no. 12, pp. 51–
57, 2006.
[13] J. R. P. Mähönen, M. Petrova and M. Wellens, “Cognitive wireless networks: Your network just became
a teenager,” in Proc. of the INFOCOM, San Francisco, CA, pp. 23–29, 2006.
[14] C. Fortuna and M. Mohorcic, “Trends in the development of communication networks: Cognitive
networks,” Computer Networks, vol. 53, no. 9, pp. 1354–1376, 2009.
[15] A. Zanella, N. Bui, A. P. Castellani, L. Vangelista and M. Zorzi, “Internet of Things for smart cities,”
IEEE Internet of Things Journal, vol. 1, no. 1, pp. 22–32, 2014.
CMC, 2021, vol.67, no.1 1283

[16] R. E. Hall, B. Bowerman, J. Braverman, J. Taylor, H. Todosow et al., “The vision of a smart
city; Brookhaven national lab,” in 2nd International Life Extension Technology Workshop, Paris, France,
pp. 1–6, 2000.
[17] T. Bakıcı, E. Almirall and J. Wareham, “A smart city initiative: The case of barcelona,” Journal of the
Knowledge Economy, vol. 4, no. 2, pp. 135–148, 2013.
[18] R. Giffinger and H. Gudrun, “Smart cities ranking: An effective instrument for the positioning of the
cities,” ACE: Architecture, City and Environment, vol. 4, no. 12, pp. 7–26, 2010.
[19] M. Batty, K. W. Axhausen, F. Giannotti, A. Pozdnoukhov, A. Bazzani et al., “Smart cities of the
future,” European Physical Journal Special Topics, vol. 214, no. 1, pp. 481–518, 2012.
[20] M. Wang, Y. Cui, X. Wang, S. Xiao and J. Jiang, “Machine learning for networking: Workflow,
advances and opportunities,” IEEE Network, vol. 32, no. 2, pp. 92–99, 2018.
[21] M. Usama, J. Qadir, A. Raza, H. Arif, K. L. A. Yau et al., “Unsupervised machine learning for net-
working: Techniques, applications and research challenges,” IEEE Access, vol. 7, pp. 65579–65615, 2019.
[22] G. Xu, Y. Mu and J. Liu, “Inclusion of artificial intelligence in communication networks and services,”
ITU Journal: ICT Discoveries, vol. Special, no. 1, pp. 1–6, 2017.
[23] A. Patcha and J. M. Park, “An overview of anomaly detection techniques: Existing solutions and latest
technological trends,” Computer Networks, vol. 51, no. 12, pp. 3448–3470, 2007.
[24] T. T. T. Nguyen and G. Armitage, “A survey of techniques for internet traffic classification using
machine learning,” IEEE Communicatons Surveys & Tutorials, vol. 10, no. 4, pp. 56–76, 2008.
[25] M. Bkassiny, Y. Li and S. K. Jayaweera, “A survey on machine- learning techniques in cognitive radios,”
IEEE Communications Surveys & Tutorials, vol. 15, no. 3, pp. 1136–1159, 2013.
[26] M. A. Alsheikh, S. Lin, D. Niyato and H. P. Tan, “Machine learning in wireless sensor networks:
Algorithms, strategies, and applications,” IEEE Communications Surveys & Tutorials, vol. 16, no. 4,
pp. 1996–2018, 2014.
[27] X. Wang, X. Li and V. C. M. Leung, “Artificial intelligence-based techniques for emerging het-
erogeneous network: State of the arts, opportunities, and challenges,” IEEE Access, vol. 3, no. 3,
pp. 1379–1391, 2015.
[28] A. L. Buczak and E. Guven, “A survey of data mining and machine learning methods for cyber
security intrusion detection,” IEEE Communications Surveys & Tutorials, vol. 18, no. 2, pp. 1153–
1176, 2016.
[29] P. V. Klaine, M. A. Imran, O. Onireti and R. D. Souza, “A survey of machine learning techniques
applied to self organizing cellular networks,” IEEE Communications Surveys & Tutorials, vol. 19, no. 4,
pp. 2392–2431, 2017.
[30] Z. M. Fadlullah, F. Tang, B. Mao, N. Kato, O. Akashi et al., “State-of-the-art deep learning:
Evolving machine intelligence toward tomorrow’s intelligent network traffic control systems,” IEEE
Communications Surveys & Tutorials, vol. 19, no. 4, pp. 2432–2455, 2017.
[31] E. Hodo, X. Bellekens, A. Hamilton, C. Tachtatzis and R. Atkinson, “Shallow and deep networks
intrusion detection system: A taxonomy and survey,” arXiv preprint arXiv, vol. 2017, pp. 1–15, 2017.
[32] X. Zhou, M. Sun, G. Y. Li and B. H. Juang, “Machine learning and cognitive technology for intelligent
wireless networks,” Information Theory, vol. 2107, pp. 1–53, 2017.
[33] M. Chen, U. Challita, W. Saad, C. Yin and M. Debbah, “Machine learning for wireless networks with
artificial intelligence: A tutorial on neural networks,” arXiv preprint arXiv, vol. 2017, pp. 1–8, 2017.
[34] B. Tang, Z. Chen, G. Hefferman, S. Pei, T. Wei et al., “Incorporating intelligence in fog computing for
big data analysis in smart cities,” IEEE Transactions on Industrial Informatics, vol. 13, no. 5, pp. 2140–
2150, 2017.
[35] S. Feng, P. Setoodeh and S. Haykin, “Smart home: Cognitive interactive people-centric Internet of
Things,” IEEE Communication Magzine, vol. 55, no. 2, pp. 34–39, 2017.
[36] Q. Wu, G. Ding, Y. Xu, S. Feng, Z. Du et al., “Cognitive Internet of Things: A new paradigm beyond
connection,” IEEE Internet of Things, vol. 1, no. 2, pp. 129–143, 2014.
1284 CMC, 2021, vol.67, no.1

[37] P. Vlacheas, R. Giaffreda, V. Stavroulaki, D. Kelaidonis, V. Foteinos et al., “Enabling smart cities
through a cognitive management framework for the Internet of Things,” IEEE Communications
Magazine, vol. 51, no. 6, pp. 102–111, 2013.
[38] M. Moore, P. Gould and B. S. Keary, “Global urbanization and impact on health,” International
Journal of Hygiene & Environmental Health, vol. 206, no. 3, pp. 269–278, 2003.
[39] P. Hall, “Creative cities and economic development,” Urban Studies, vol. 37, no. 4, pp. 639–649, 2000.
[40] R. P. Dameri, “Defining an evaluation framework for digital cities implementation,” in Int. Conf. on
Information Society (i-Society 2012), London, pp. 466–470, 2012.
[41] R. Kitchin, “The real-time city? Big data and smart urbanism,” GeoJournal, vol. 79, no. 1,
pp. 1–14, 2014.
[42] S. Scott-Hayward, P. K. Chouhan, B. Fraser, D. Lake, S. Sezer et al., “Are we ready for SDN?
Implementation challenges for software-defined networks,” IEEE Communincation Magazine, vol. 51,
no. 7, pp. 36–43, 2013.
[43] F. Francois and E. Gelenbe, “Optimizing secure SDN-enabled inter- data centre overlay networks
through cognitive routing,” in IEEE 24th Int. Symp. on Modeling, Analysis & Simulation of Computer &
Telecommunication Systems, London, pp. 283–288, 2016.
[44] E. Raimundo-Neto, J. R. G. da Rosa, M. A. F. Casaroli, I. Feliciano da Costa, A. M. Alberti et al.,
“Implementation of an optical-wireless network with spectrum sensing and dynamic resource alloca-
tion using optically controlled reconfigurable antennasg optically controlled reconfigurable antennas,”
International Journal of Antennas & Propagation, vol. 2014, pp. 1–14, 2014.
[45] B. Shan, H. Ji and Y. Li, “Centralized compressed sensing with structurally random matrix in cognitive
WLAN over fiber,” in Wireless Communications & Networking Conference, 2013 IEEE, Shanghai, China,
pp. 106–111, 2013.
[46] Y. Li, H. Ji, X. Li and V. Leung, “Dynamic channel selection with reinforcement learning for cognitive
WLAN over fiber,” International Journal of Communication Systems, vol. 25, no. 8, pp. 1077–1090, 2012.
[47] R. Urgaonkar and M. J. Neely, “Opportunistic cooperation in cognitive femtocell networks,” IEEE
Journal on Selected Areas in Communications, vol. 30, no. 3, pp. 607–616, 2012.
[48] S. Al-Rubaye, A. Al-Dulaimi and J. Cosmas, “Cognitive femtocell,” IEEE Vehicular Technology Maga-
zine, vol. 6, no. 1, pp. 44–51, 2011.
[49] Q. Yan, F. R. Yu, Q. Gong and J. Li, “Software-defined networking and distributed denial of service
(DDoS) attacks in cloud com- puting environments: A survey, some research issues, and challenges,”
IEEE Communications Surveys & Tutorials, vol. 18, no. 1, pp. 602–622, 2015.
[50] B. Yuan, D. Zou, S. Yu, H. Jin, W. Qiang et al., “Defending against flow table overloading attack in
software-defined networks,” IEEE Transactions on Services Computing, vol. 12, no. 2, pp. 231–246, 2016.
[51] S. Dev, B. Wen, Y. H. Lee and S. Winkler, “Ground-based image analysis: A tutorial on machine-
learning techniques and applications,” IEEE Geoscience & Remote Sensing Magazine, vol. 4, no. 2,
pp. 79–93, 2016.
[52] C. S. Nwosu, S. Dev, P. Bhardwaj, B. Veeravalli and D. John, “Predicting stroke from electronic health
records,” in Proc. 41st Annual Int. Conf. of the IEEE Engineering in Medicine and Biology Society, Berlin,
Germany, pp. 5704–5707, 2019.
[53] S. Abbas, M. A. Khan, L. E. Falcon-Morales, A. Rehman, Y. Saeed et al., “Modeling, simulation and
optimization of power plant energy sustainability for IoT enabled smart cities empowered with deep
extreme learning machine,” IEEE Access, vol. 8, no. 1, pp. 39982–39997, 2020.
[54] A. Rehman, A. Athar, M. A. Khan, S. Abbas, A. Fatima et al., “Modelling, simulation, and optimiza-
tion of diabetes type II prediction using deep extreme learning machine,” Journal of Ambient Intelligence
and Smart Environments, vol. 12, no. 2, pp. 125–138, 2020.
[55] M. A. Khan, S. Abbas, K. M. Khan, M. A. Ghamdi, A. Rehman et al., “Intelligent forecasting model
of COVID-19 novel coronavirus outbreak empowered with deep extreme learning machine,” Computers,
Materials & Continua, vol. 64, no. 3, pp. 1329–1342, 2020.
CMC, 2021, vol.67, no.1 1285

[56] Kaggle 2020. [Online]. Available: https://www.kaggle.com/hassan06/nslkdd.

[57] Y. J. Li, X. B. Li and O. Yoshie, “Traffic engineering framework with machine learning based meta-
layer in software-defined networks,” in Proc. IEEE ICNIDC’14, Beijing, China, pp. 121–125, 2014.
[58] C. X. Cui and Y. B. Xu, “Research on load balance method in sdn,” International Journal of Grid &
Distributed Computing, vol. 9, no. 1, pp. 25–36, 2016.