Journal of Science Engineering Technology and Management Science

Volume 01, Issue 01, December 2024

www.jsetms.com

Wireless Networks with Machine Learning Routing Enabled

Swathi Priya1 and Swetha Reddy2
1,2
Department of Electronics and Communication Engineering, Usha Rama Engineering college, Vijayawada
Corresponding Author: [email protected]

To Cite this Article

Swathi Priya and Swetha Reddy, “Wireless Networks with Machine Learning Routing Enabled”, Journal of Science
Engineering Technology and Management Science, Vol 01, Issue 01, December 2024, pp: 22-27

Abstract: Based on the network's modifications and the type of data it perceives; the machine learning system was
specifically created to save energy and extend sensor life. Additionally, by combining similar types of data that are
sensed inside the network coverage region, the sensor network is created using machine learning and the BAT
computational approach to minimise the network under redundancy. Additionally, the network uses very little energy
from the sensors, therefore feature sets in fuzzy-neuron machine learning mode are employed to choose neighbours.
Furthermore, the packet speed sensed and the detecting intervals between packets are used to calculate the energy
used by the sensor in the network. Additionally, the neural network is used to open and aggregate the data. The
approach designed to reduce the aggregation method's energy use. The same data is then aggregated once that is
finished (removing the noise from the data). This lowers energy consumption and increases the use of network
resources. Furthermore, a routing path that optimises during the routing path development process is created using
BAT computation, resulting in a consistent and energy-efficient path.
Keywords: BAT routing, fuzzy neural network, aggregation, redundancy, and sensor network
_____________________________________________________________________________________
I. Introduction
The environment of the sensing region determines the variation of the data group, which can be formed by
turning the sensors into groups. To be more precise, the data ought to be arranged according to its characteristics. If
not, a collection of disparate sensors could create an impractical cluster set. The fuzzy inference system provides the
best solution to this issue. The moment the data is sensed, its location, the distance between those two sensors, and
other details are all calculated. Furthermore, we get precise data reports from the neural network, compare them with
the energy prices of each one, and classify the data according to its features and variations. We reduce a predetermined
set of the hidden layers as the inputs. Next, we adjust the weights in this minimised set by giving the hidden layer's
input beginning values. Lastly, the machine learning approach speeds up the hunt for answers. This optimisation and
routing selection for data transfers are further carried out using the computing approach. This computation uses the
approach of an intelligent bird that can search the entire world. Rather, it aims to improve the search and machine
input learning. The sensor takes into account the amount of energy left, the transmission distance, and other variables
while choosing this path. If so, the sensors installed on it will increase its maximum coverage area, conserve network
resources, and prolong the sensor's lifespan as it takes
This paper's Section 2 describes the specifics of earlier studies on modifications to sensor network routing.
Section 3 goes into great detail about the suggested SFNB sensor data optimisation using fuzzy neurones and BAT
computation-based machine learning routing in a wireless network. The SFNB results and discussions are presented
in section 4, the last portion. This report wraps up and makes recommendations for more research.

II. Literature Survey

Electrical power operation depends critically on data from wireless sensor networks used for electrical load
monitoring. This results in lower operational costs and increased energy supply reliability [4], which necessitate
precise load estimates to guarantee appropriate power system operation and planning. The fascinating region can be
monitored using WSNS through multi-hop communication. Coverage is the main parameter used to assess monitoring
capacity. As a result, it also ensured connectivity, allowing the BS to receive all sensed data for processing in the
future [1]. Third, we believe that with a low overhead execution environment for those primitives, the underlying
architecture should allow rich extensions of basic macro program primitives using the application domain or the
system's operational requirements. The variety WSNs are expected to be used for national security and surveillance,
environmental and habitat monitoring, and almost anything in between. In fact, the combination of wireless

Published by: Surya Publishers www.jsetms.com 22 | Page

 Wireless Networks with Machine Learning Routing Enabled

communications and microelectronics has already made WSNs the much-anticipated solution for complex, large-scale
information processing and decision-making activities. There may be variations in the size and cost of specific sensor
nodes. WSN physical characteristics could resemble those of a multi-hop wireless mesh network or a star network.
The former was required since the virtual sensor predictions were influenced by atypical environmental conditions
and error-prone sensor readings. By enabling these machine learning-based virtual sensors to replace the broken ones,
the suggested approach helps to strengthen WSNs. Due to energy and bandwidth limitations, it is simply not feasible
for a sensor to send all of the data back to a base station for processing and drawing conclusions. This is one of the
issues that arises when a high number of variables needs to be tracked.
Consequently, it is necessary to apply machine learning techniques to combat the WSN [2]. During the
clustering process, the network is split up into interconnecting substructures known as clusters. Every cluster in the
substructure has a cluster leader who serves as its coordinator. The cluster head serves as a conduit for data between
the nodes. CHs communicate with other CHs through gateway nodes [7]. We created a real-world monitor that can do
deep learning-based analysis before the deadline in a time-constrained setting. We report on simulation utilising
published data and deep learning performance under time limitations to verify the effectiveness of the suggested
method.
An arithmetic algorithm called a neural network may learn a complex mapping from input to output. The act
of gathering and analysing sensor data in order to reduce network data transmission is known as data aggregation [3].
A WSN (WSN) can sense and measure a wide range of physical or environmental conditions, including multimedia,
infrared energy from objects that can be converted to their temperature, pollutant levels, ultrasound in medical
imaging, vibration, security and surveillance, agriculture, and many more. Sensor node data transfer uses a lot of
energy. Thus, the network's lifetime can be extended and energy waste can be minimised by minimising superfluous
communication [5]. However, sensor networks have a finite lifespan. Therefore, it is necessary to revitalise the sensor
network by making more nodes accessible. Heterogeneity in node energy is produced because more energy will be
supplied than nodes that are currently using it. This TDMA scheduling results in a network that uses less energy and
has a longer lifespan. To send the data back to the BS, the CH additionally filters out redundant information from the
received data and fuses it further. For long-term deployments, sensor networks at remote locations, such as those used
for animal tracking, need to be extremely resilient to harsh environments.

III. Proposed Work

Figure no 1: Groups formed by Data

These events are the result of the network's delivery of a path selection system, which we intend to build up
with both local and global forwarder selection. The BAT computation will decide the route on its own. Thus, the
specified criteria finish the paving route in a very short amount of time using the most reliable local and worldwide
searches. In the first case, packets are sent through the network settings once all the nodes are arranged at the start of

23 | Page
 Wireless Networks with Machine Learning Routing Enabled

the network. First, using the F(I) and neurone system, we categorised the Dg as a cluster in this suggested way. This
made it possible to generate additional D(G) to identify particular data features. We classified using a few criteria,
such as node position and sensing counts, sensing time, ID, and the network used these few characteristics to sort. To
get a more accurate D(G), all of these parameters are then processed F(I) and the fuzzy set's output is passed to the
neural network. Additionally, each sensor's neighbour sets are chosen to be the node with the shortest distance and the
highest energy towards the destination. D(G) will be changed in accordance with changes in the environment and node
position, which will then update the network and rebuild the data group. The continuous monitoring sensor predicted
the incoming data status. The D(G) altered. Nodes can calculate to notice in the suggested approach. In order to
determine the routing path, the BAT computation forecasts the forwarder nodes' local and global quality.

Base Station Forward node End node

Base Station Forward node End node

Figure no 2: Selection of groups based on data

24 | Page
 Wireless Networks with Machine Learning Routing Enabled

IV. Conclusion and Conversation

In this network, these have been modified to 400 nodes and tested. We randomly dropped the nodes and set
the network area to be between 200 and 200. Random data nodes sense the application as it moves. Its performances
are shown in graph form below.

Lifetime
10150
10100
10050 DCHH
S
10000
ELM
9950
9900
9850
1 2 3 4 5
Dataset

Figure no 3: Lifetime vs Dataset

The quantity of life that nodes in a network can utilise and extend based on energy is known as the energy
network lifetime. Consequently, a network with nodes that have a longer lifespan will function better. Better network
routing may be possible for the longer energy use. The suggested model, SFNB, has a longer lifespan than other
ACNM, ELM, and DCHHS methods, as Figure 3. illustrates. This neighbour set is created utilising unique information
about the data that is perceived in the network because it is a sensor network. The data is then used to choose the best
routing. This prolongs the lifespan, lowers energy losses, and speeds up data transfer.

Packet Delivery Ratio

100

95

90 DCHHS
ELM
85
ACNM
80 SFNB

75

70
100 125 150 175 200

Node

Figure no 4: Packet Delivery Ratio vs Node

The ratio of packets transported to their destination is known as the packet delivery ratio. If the
communication is effective, the networks will become more and more successful. Otherwise, the proportion will drop.
The primary causes of this are neighbour selection and routing. Other procedures' PDR is lower than ours in SFNB,
as seen in Figure 4. The neighbour set is used to create the data groups, after which the proper routing is chosen. This
selects numerous network parameters, such as energy, data type, distance, etc., after which the neighbours are chosen

25 | Page
 Wireless Networks with Machine Learning Routing Enabled

based on the properties of the data. Additionally, we use BAT optimisation to generate routing. As a result, a reliable
network has been established, increasing the proportion of data delivery.

Delay

0.8
0.7
0.6
DCHHS
0.5
ELM
0.4
ACNM
0.3
SFNB
0.2
0.1
0
100 125 150 175 200

Simulation Time

Figure no 5: Delay vs Simulation Time

Network latency is the term used to describe transmission time delay. How long this last depends on how
stable the network is. Numerous factors, including the wireless channel's characteristics, the nodes' connecting times,
and the packet buffering time, affect this network delay. We can presume that the network protocol is operating with
or very near 100% correctness if the delay is brief. After dividing a brief feature by its feature set, it is sorted and sent
into the SFNB algorithm, which chooses the neighbours with accuracy. Thus, it may be concluded that the suggested
system's performance has improved (Figure 5).

Throughput

90000
80000
70000
60000 DCHHS
50000 ELM
40000 ACNM
30000 SFNB
20000
10000
0
100 125 150 175 200

Node

Fig 6: Throughput vs Node

26 | Page
 Wireless Networks with Machine Learning Routing Enabled

The quantity of packet bits that reach their destination is known as throughput. The packets will reach their
destination rapidly if the network operates consistently. After then, its bit count will also rise. One of the characteristics
of network measurements is throughput. As seen in Figure.6, the suggested approach SFNB has a high throughput in
this case. These differences are the result of optimal routing selection and group formation. This involves doing BAT,
fuzzy, and neural computations to improve the wireless communication sensor network.

V. Final Thoughts
The network must adapt when we receive data that needs to be sensed somewhere in the network and sent to
a different location. For the sensors to meet these requirements, the networks need the right protocol. The path is then
selected by the SFNB data group that was created here. The sensors have generated data groups using the fuzzy
neurone method, and the routing path has been chosen using BAT computations. To create accurate groups and split
the data characteristics into data groups, a variety of factors are used. It won't be confusing when the data is aggregated.
Similarly, BAT calculation determines how to transfer the data according to the recipients. It lowers energy costs by
choosing the best forwarders, both local and international, and sending data through them. In the future, this network
will offer safe ways to route data so that it can be sent by secure channels.

References
[1]. Vijay muni, Raja Sekhar and Srinivasa varma, “Implementation of AI in various sectors of wireless networks”, Gurur nank institutes,
Hyderabad, conference proceedings, vol 5, issue 2, June 2012
[2]. Manohar singh, Venkata ramana and sekhar suman, “ Case study on machine learning in wireless networks in sensor networks”, VNR
Vignan institutes, Hyderabad, conference proceeding, vol 7, issue 3, Septembe2 2021.
[3]. Manoj, Mohan kumar and Rajkumar,” Implementation of LAN and WAN in different scenarios of relay networks”, JNTUH, Hyderabad
IEEE Standards, vol3, Issue 5, pp 46-54.
[4]. Kavitha Srinivas reddy , Manohar joshi and Prasad Raju, “Optimization of Neural networks in machine learning using AI in Networks”,
Gokaraju, Springer conference proceedings, vol 6, issue 7.
[5]. Niharika, Bindu Priya and Madhvi Reddy, " Energy Control in Wireless Sensor Systems using Machine Learning”, JNTUH, IEEE
conference proceedings, vol 4, issue 2, pp:54-62
[6]. Srikanth Reddy, Surya and Md Abdullah, “ Study of MIMO in wireless relay networks in space frequency codes”, DNR college, Springer
lecture notes, 2022
[7]. Ramesh kumar, Daniel, Mohan Reddy and Weing shah, “Case study of Distibuted space time and space frequency codes to acieve
cooperative diversity in relay networks”, KL University, Springer lecture notes 2022

27 | Page