IoT Attack Detection Using Artificial

Neural Network
Animesh Srivastava [1] Sant Kumar Maurya [2] Parveen Kumar Saini [3]
Assistant Professor [1] Assistant Professor [2] Assistant Professor [3]
Department of Computer Science & Department of AIT-CSE, Chandigarh Department of CSE, Chandigarh
Engineering University, Gharuan, Mohali, Punjab [2] University, Gharuan, Mohali, Punjab [3]
Graphic Era Hill University, [email protected] [2] [email protected] [3]
Dehradun
[email protected] [1]

Shiv Ashish Dhondiyal [4]

Department of Computer Science &
Engineering [4]
Graphic Era (Deemed to be
University), Dehradun [4]
[email protected] [4]

Abstract

to detect and identify this change. A cyberattack,

As billions of devices are now connected to the
internet, IoT has become an integral part of modern and more specifically a Distributed Denial of
life. However, hackers are constantly developing new Service (DDoS) attack, could be to blame [4][5].
methods of exploiting IoT devices, leaving them open
to security breaches. There are many applications for Artificial neural networks (ANNs) can be used in
artificial neural networks (ANNs) in the detection and many different ways to detect attacks, one of which
analysis of IoT attacks. The structure and operation is to identify compromised Internet of Things (IoT)
of ANNs are inspired by the human brain. This gadgets. An artificial neural network (ANN) could
enables them to mine data for hidden patterns. This
makes them ideal for discovering IoT attacks, which be trained to detect and label new network
frequently involve convoluted data patterns. To learn connections, for instance. The observed activity
how to identify patterns in IoT traffic data, a deep may indicate an unauthorized attempt to log in,
neural network is employed. We put our method which could be a brute-force attack [6][7].
through its paces using real-world IoT traffic data.
Our findings demonstrate that our approach is highly When it comes to protecting IoT networks from
effective at detecting IoT attacks. intrusion, ANN has many advantages over more
traditional methods.
Keywords: IoT, attack detection, artificial neural networks,
deep learning To begin with, the method shows impressive
effectiveness in efficiently learning from large
1.0 Introduction datasets. The importance of this topic lies in the
fact that the massive amounts of data generated by
When it comes to protecting the IoT from cyber IoT devices threaten to swamp more traditional
criminals, artificial neural networks (ANNs) are a approaches to detecting attacks.
useful form of machine learning. The structure and
operation of ANNs are inspired by the human For another, ANN can learn complicated patterns.
brain. This enables them to mine massive datasets The relevance of this issue lies in the fact that
for hidden insights. Since attacks on the IoT attacks on the IoT are typically highly
frequently involve convoluted data patterns, these sophisticated, making it difficult for traditional
tools are ideal for detecting them. detection methods to pick them up.

Finding and fixing security flaws in the IoT can be Third, ANN demonstrates the ability to adapt and
accomplished in many ways using Artificial Neural respond to new threats effectively. The significance
Networks (ANNs). Artificial Neural Networks of this issue lies in the ever-changing nature of
(ANNs) could be used to identify illogical traffic attacks aimed at the IoT.
patterns. The traffic going to a specific IoT device
may suddenly increase, and an ANN can be trained
2.0 Related Work

IoT attack detection using ANNs, offering different 2.1 GAP ANALYSIS
techniques, datasets, and insights to improve IoT
security through machine learning. have looked Since ANNs can learn and discover patterns in
into different ANN architectures, data data, they have been selected as a useful tool for
preprocessing methods, and network traffic discovering attacks in IoT systems. Since attacks
analysis strategies to make IoT systems more on the IoT frequently involve convoluted data
secure against attacks [1][2]. patterns, these tools are ideal for detecting them.

Numerous individuals have worked diligently to • ANNs can learn patterns that aren't simple. This is
detect and prevent Internet of Things (IoT)-related important because attacks on IoT devices are often
attacks. A significant amount of research has been very smart and can avoid being found in the usual
conducted on IoT devices and network security. As ways.
new vulnerabilities and countermeasures are • ANNs can change so they can handle new
discovered, security for IoT devices is a rapidly problems. This is important because attacks on IoT
developing field. [8] [9]. are getting better and better all the time.
Internet of Things (IoT) cyberattacks can wreak • A lot of people can use ANNs. This means that
havoc on the economy and people's health by they can be used to find attacks on a lot of Internet
disrupting production lines, manufacturing of Things devices.
processes, and supply chains. A system that
prevents attacks employs Graph Neural Networks 2.2 MOTIVATION
and Reinforcement Learning to analyze alerts and
The data produced by IoT (Internet of Things)
reroute sensitive traffic away from compromised
devices is substantial. The provided data can be
network paths. The detection and rerouting
used to train ANNs to identify patterns that can be
schemes are demonstrated by experimental results.
used to foresee attacks. Because of the high level of
[12][13].
sophistication required to launch an IoT attack,
IoT systems are susceptible to numerous types of conventional detection methods are often
attacks, such as DoS and DDoS. Determining and ineffective. Because of their ability to learn
preventing denial-of-service/denial-of-service complex patterns, artificial neural networks
attacks are proposed as two distinct components of (ANNs) are excellent at detecting these assaults.
a brand-new architecture. On the Bot-IoT dataset, The nature of threats posed by the IoT is constantly
the proposed detection component, a "Looking- shifting. Updates to ANNs give them the ability to
Back" multi-class classifier, is evaluated. [3][14] pick up on novel patterns, expanding their
[15]. usefulness. They are adept at fending off novel
threats because of their adaptability. The ability to
 The choice of ANN algorithm for IoT expand is a unique feature of artificial neural
attack detection will depend on the networks (ANNs). This allows them to detect and
specific application. thwart attacks across a wide range of IoT gadgets.
 Some of the challenges in using ANNs for Furthermore, artificial neural networks (ANNs)
IoT attack detection include the lack of appear to be a good way to detect attacks on the
labelled data, the dynamic nature of IoT IoT because of the following:
traffic, and the resource constraints of IoT
devices. Effective: ANNs are good at finding a variety of
 Despite these challenges, ANNs have the IoT attacks, such as DDoS attacks, malware
potential to be a powerful tool for IoT infections, and unauthorized access.
attack detection.
Scalable: ANNs can be made bigger so that they
Using artificial neural networks (ANNs) to find can find attacks on a lot of IoT devices.
attacks on the Internet of Things (IoT) is a hard
Adaptable: ANNs can be updated to learn new
task that requires training machine learning models
patterns, which makes them able to deal with new
to find patterns and outliers in IoT data streams
threats.
[10][11].
3.0 PROPOSED FRAMEWORK
Data collection. Obtaining information from IoT
gadgets is the first step. Sensors, actuators, and
other hardware can be used to gather this
information. The information gathered from the IoT
devices should be indicative of their typical
operation.

Feature extraction. The gathered data must then be

processed so that features can be extracted. (Such
as volume, time of day, origin and destination
addresses, and so on)

Pre-processing of data. Before the ANN can make

use of the extracted features, they may require
some pre-processing. Possible strategies include
dimensionality reduction, feature selection, and
normalization.

Model training. The data is cleaned and prepared

for training the ANN. The ANN is trained to
recognize safe and malicious connections.

Threat identification. Once the ANN has been

trained, it can be used to make real-time security
assessments. The ANN determines whether or not Figure 1: Anomaly detection
incoming traffic represents an attack.
Here are the steps involved in IoT attack detection
DDoS Detection and prevention using random forests: To do this,
we need to collect data from IoT devices. This
The following conditions will be used for the
includes information gleaned from networks,
analysis in this session:
devices, and sensors. It's important to gather this
1. The typical amount of data exchanged by information safely and reliably Following data
a node. collection, feature extraction is necessary for attack
detection. The length, transmission rate, and
2. The rate of transmission used for each pair protocol of a packet are all examples. Features
of nodes. should be chosen with care to ensure they are easily
3. The typical lag in message delivery across accessible and useful. Get the model training
all nodes underway. Building an intrusion-detection
machine-learning model is the next step. ANN an
4.0 FRAMEWORK efficient and extensible algorithm, is commonly
used in this context. Both secure and insecure data
The ANN machine learning algorithm is proposed flows were used in the model's training set. The
as part of a framework for protecting IoT devices trained model can then be used to detect active
from cyberattacks. It does this by building multiple attacks. The analysis performed by the model will
decision trees and then taking an average of the determine the data's quality. An alert will be sent
results. This could lead to a less-vulnerable model out if the traffic quality is poor. Once an attack has
that is more accurate. been spotted, we need to put a stop to it
In Figure 1: we have discussed anomaly detection immediately. It is possible to prevent the malicious
in the IoT. data or the compromised computer from accessing
the network. A framework like Random Forest
makes it simple to keep malicious actors from
disrupting the Internet of Things. Methods and
specifics of implementation may change based on
the specifics of the situation. Here are some of the
benefits of using ANN for IoT attack detection:

 It is a robust and scalable algorithm.

  It can be trained on a variety of datasets. stage of the attack, thus preventing major
disruptions to the network.
 It is relatively easy to interpret.
6.0 Conclusion
When it comes to identifying malicious activity in
the IoT, ANN is a promising method. This We showed how to identify DDoS/DOS attacks on
algorithm is robust and extensible, and it can be an IoT network using a neural network-based
trained using data from many sources. intrusion detection method. The detection was
based on figuring out what was normal and what
was a threat. The ANN model was tested on a
5.0 Analysis of Results
simulated IoT network, which
This section provides an evaluation of the showed 99.2% accuracy. One major advantage of
performance. The network was trained with 2310 using ANNs for IoT attack detection is their ability
samples, and validated with 490 samples and 490 to learn and adapt to new attack patterns. This is
test samples. essential because the cybersecurity industry must
constantly adjust to new threats. The ability of
Scenario: An IoT device is being attacked, and a ANNs to detect attacks that would otherwise go
TCP flooding attack is found. With linear SVM, we undetected is one of their main advantages. This is
see a 93% level of accuracy. When non-linear SVM because ANNs are capable of detecting and
is used, on the other hand, 97.8%, Random Forest responding to even the most minute changes in
96%, In this our proposed ANN has an accuracy of traffic patterns that may indicate an attack. The
99.2%. experimental results demonstrate that the simulated
IoT environment can successfully detect various
Table 1: Compare with other Methods
DDoS/DoS attacks, with an accuracy of 99.2%. We
Method Accurac plan to explore additional, more advanced neural
y network techniques, such as recurrent and
Proposed Artificial Neural Network 99.2% convolutional networks, in the near future.
non-linear SVM 97.8%
linear SVM 93%
Random Forest [4] 96%
Autoencoder [4] 70% Reference
1. S. S. Bhunia and M. Gurusamy, "Dynamic attack detection
and mitigation in IoT using SDN," 2017 27th International
120.00% Telecommunication Networks and Applications Conference
100.00% (ITNAC), Melbourne, VIC, Australia, 2017, pp. 1-6, doi:
80.00% 10.1109/ATNAC.2017.8215418.
60.00%
40.00% 2. E. Gelenbe et al., "IoT Network Attack Detection and
20.00% Mitigation," 2020 9th Mediterranean Conference on Embedded
0.00% Computing (MECO), Budva, Montenegro, 2020, pp. 1-6, doi:
Accuracy

k
or M M [4] [4] 10.1109/MECO49872.2020.9134241.
tw SV r SV est er
r d
Ne ea ea or co 3. Alaeddine Mihoub, Ouissem Ben Fredj, Omar Cheikhrouhou,
ral -lin lin m F oen Abdelouahid Derhab, Moez Krichen, Denial of service attack
eu on o t
lN n nd Au detection and mitigation for internet of things using looking-
ic a Ra back-enabled machine learning techniques, Computers &
tifi
Ar Electrical Engineering, Volume 98, 2022, 107716, ISSN 0045-
d 7906, https://doi.org/10.1016/j.compeleceng.2022.107716.
se
opo
Pr 4. E. Gelenbe et al., "IoT Network Attack Detection and
Mitigation," 2020 9th Mediterranean Conference on Embedded
Computing (MECO), Budva, Montenegro, 2020, pp. 1-6, doi:
Figure 2: Compare with other Methods 10.1109/MECO49872.2020.9134241
Overall, the neural network model has a 5 A. Srivastava and A. Kumar, "A back propagation NN to
classification accuracy of 99.2%. This model optimize the IoT network," 2022 International Conference on
proves the deployed ANN algorithm can identify Computer Communication and Informatics (ICCCI), 2022, pp.
1-4, doi: 10.1109/ICCCI54379.2022.9740861.
DDoS/DoS attacks against benign IoT network
traffic. In addition, it aids in bettering network 6 Singh, Gulbir & Kumar, Gautam & Bhatnagar, Vivek &
stability by alerting the response team at an early Srivastava, Animesh & Jyoti, Kumari. (2019). POLLUTION
MANAGEMENT THROUGH INTERNET OF THINGS: A
SUBSTANTIAL SOLUTION FOR SOCIETY. Humanities &
Social Sciences Reviews. 7. 1231-1237.
10.18510/hssr.2019.75162.

7. A. Srivastava and A. Kumar, "A Review on Authentication

Protocol and ECC in IOT," 2021 International Conference on
Advance Computing and Innovative Technologies in
Engineering (ICACITE), 2021, pp. 312-319, doi:
10.1109/ICACITE51222.2021.9404766.

8. Leme, A. Medeiros, G. Srivastava, J. Crichigno and R. Filho,

"Secure Cattle Stock Infrastructure for the Internet of Things
using Blockchain," 2020 43rd International Conference on
Telecommunications and Signal Processing (TSP), Milan, Italy,
2020, pp. 337-341, doi: 10.1109/TSP49548.2020.9163494

9 Srivastava, A., Kumar, A. (2022). A Review of Network

Optimization on the Internet of Things. In: Saini, H.S., Sayal,
R., Govardhan, A., Buyya, R. (eds) Innovations in Computer
Science and Engineering. Lecture Notes in Networks and
Systems, vol 385. Springer, Singapore.
https://doi.org/10.1007/978-981-16-8987-1_6

10 Srivastava, A.., & Kumar, A. (2023). Secure Authentication

Scheme for the Internet of Things. International Journal on
Recent and Innovation Trends in Computing and
Communication, 11(4s), 182–192.
https://doi.org/10.17762/ijritcc.v11i4s.6368

11 Mishra, S. (2021). Detection and mitigation of attacks in

SDN-based IoT network using SVM. International Journal of
Computer Applications in Technology, 65(3), 270-281.

12. A. Chauhan, I. Kumar, C. Bhatt and A. Agnihotri,

"Computer aided Breast lesions classification system using
digitized fine needle aspirate images," 2022 OPJU International
Technology Conference on Emerging Technologies for
Sustainable Development (OTCON), Raigarh, Chhattisgarh,
India, 2023, pp. 1-6, doi:
10.1109/OTCON56053.2023.10113974.

13. A. Giri, D. S. Bisht, A. Chauhan and I. Kumar,

"Computerized Face Mask Detection System Using Deep CNN
and Transfer Learning," 2023 International Conference on
Device Intelligence, Computing and Communication
Technologies, (DICCT), Dehradun, India, 2023, pp. 457-462,
doi: 10.1109/DICCT56244.2023.10110187.

14. S. Banerjee, D. Sethia, T. Mittal, U. Arora and A. Chauhan,

"Secure sensor node with Raspberry Pi," IMPACT-2013,
Aligarh, India, 2013, pp. 26-30, doi:
10.1109/MSPCT.2013.6782081.

15. A. Srivastava, B. S. Rawat and S. K. Maurya, "A Review on

Protecting SCADA Systems from DDOS Attacks," 2023 4th
International Conference on Electronics and Sustainable
Communication Systems (ICESC), Coimbatore, India, 2023, pp.
556-561, doi: 10.1109/ICESC57686.2023.10193650.