Enhancing QoS in 5G IoT with CNN and Blockchain Security vs.
Deep Reinforcement Learning
Jeya Darshini and V. Nagaraju
Saveetha University, Chennai, Tamil Nadu, 602105, India
Keywords: 5G-Enabled Internet of Things, Novel Convolutional Neural Network, Deep Reinforcement, Blockchain
Framework, Intrusion Detection, Network Infrastructure.
Abstract: In this research, the efficiency of both Novel Convolutional Neural Network (CNN) and Deep Reinforcement
Learning (DRL) was assessed in enhancing the Quality of Service (QoS) for 5G-enabled intelligent Internet
of Things (IoT) systems. Using data from the Kaggle repository with a sample size of 5840, an 80% G power
and a 95% confidence interval were established. Two groups of 20 iterations each were divided, with Novel
Convolutional Neural Networks making up the first group, and Deep Reinforcement Learning constituting
the second. The results revealed that the accuracy of Deep Reinforcement Learning stood at 70.70%, whereas
the CNN yielded an 84.64% accuracy rate. A marked difference of 0.405 between the two groups was
observed, indicating non-significance. Therefore, it's evident that CNNs offer superior QoS accuracy over
Deep Reinforcement Learning.
1 INTRODUCTION
The advancement of intelligently-enabled Internet of
Things (IoT), powered by 5G technology, has
transformed our everyday experiences. 5G technology
facilitates faster, more dependable communication
networks, laying the foundation for the evolution of
smart wearables, healthcare systems, and other IoT
services. It offers low latency, great scalability (Qi and
Liu 2018), and a refined network infrastructure that
ensures efficient data processing and communication.
The advent of 5G also allows real-time data
transmission, enhancing communication and user
experience. This innovation supports the connection of
myriad devices to a singular network, ensuring smooth
communication. A 5G-ready IoT application (Rathore
et al. 2021) promises to optimise user satisfaction,
service quality, and network experience by connecting
a vast number of devices. To guarantee the secure
transmission of data without depending on a central
authority, blockchain technology has been
recommended. Utilising a distributed ledger system,
blockchain offers a P2P transaction platform, ensuring
data is safely, ncryptedly, and decentralisedly
recorded, verified, and exchanged (R. Pavaiyarkarasi
et al. 2022). This technology promises heightened
security and privacy for next-generation network
communication infrastructures (Deena, S. R et al
2022). Blockchain's decentralisation, security, and
anonymity (Mahapatra, S et al. 2016) have the
potential to be revolutionary for 5G-enabled IoT
networks. To evaluate the efficiency of these
innovations, it's imperative to employ Quality of
Service (QoS) metrics (Pradhan et al. 2021),
encompassing factors like accuracy, latency, and
security. Our methods incorporate QoS factors,
ensuring their effectiveness in 5G-enabled IoT
applications (Pradhan et al. 2021; Verhelst and Moons
2017). They guarantee the successful deployment of
real-time applications by providing a secure and
adaptable environment for 5G IoT devices.
Recent research has strived to enhance firewall
protection for IoT devices. As per Google Scholar,
over 15,320 publications cover this area, and IEEE
Xplore has 4,680 articles (Rathore et al. 2021).
Incorporating a blockchain security framework with
deep learning algorithms has been proposed for
heightened privacy and accuracy. Some studies
(Rathore et al. 2021; Anand and Khemchandani 2020)
have highlighted security and privacy concerns within
the fog computing system layers and suggested
solutions involving cloud and edge planes. Another
research (Sahu et al. 2021) focused on malicious
attacks in the security layer, offering a blockchain
framework to detect such attacks on 5G IoT devices
using deep learning. These researchers emphasise the
pressing need for robust security measures due to the
58
Darshini, J. and Nagaraju, V.
Enhancing QoS in 5G IoT with CNN and Blockchain Security vs. Deep Reinforcement Learning.
DOI: 10.5220/0012558800003739
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 1st International Conference on Artificial Intelligence for Internet of Things: Accelerating Innovation in Industry and Consumer Electronics (AI4IoT 2023), pages 58-63
ISBN: 978-989-758-661-3
Proceedings Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.
inherent vulnerabilities of IoT devices. The collective
application of blockchain technology and deep
learning algorithms can significantly bolster IoT
security.
Existing research indicates that combining Deep
Learning algorithms with Blockchain firewall
improves accuracy in 5G-enabled IoT devices.
However, there's a noticeable gap in identifying
threats. The Novel Convolutional Neural Network
has been identified as a potent solution to counteract
such threats. Moreover, there's an urgent need to
investigate how the proposed strategies can address
prevailing challenges in IoT systems, including
security, scalability, and performance.
2 MATERIALS AND METHODS
At the Machine Learning Lab in the Saveetha School
of Engineering, part of the Saveetha Institute of
Medical and Technical Sciences, a comprehensive
study was conducted. To enhance accuracy, each
group underwent a series of 10 iterations. The dataset
for this study was sourced from the Kaggle website.
Python served as the primary programming language
for all experimental procedures. The samples were
methodically divided, with 80% designated for
training purposes and the remaining 20% allocated
for testing. Every group comprised 24 samples. A
significance level of 0.05% and a confidence interval
of 95% were set, with G-Power set at 80%.
The mean and standard deviation for the sample
size were determined based on data collated from
multiple websites (Riaz et al. 2022).
2.1 Novel Convolutional Neural
Network
The Convolutional Neural Network (CNN) is a
flexible neural network utilised for pattern
recognition and image processing. It establishes a
structured flow for training parameters through its
input, convolution, pooling, and output layers. In the
convolution layer, an input image undergoes filtering
to derive feature maps essential for the convolution
process. Subsequently, the pooling layer down-
samples these feature maps using an activation
function, scalar weighting, and bias. One significant
advantage of a CNN is the parallel nature of its
learning, which simplifies implementation. The
computations and operations within a CNN's layers
can be articulated by equations (1), (2), (3), and (4)
(Tanwar 2021).
















(1)
Where


is the output at layer L, feature
pattern K, row x, column y of the convolutional core
which is denoted by f.
Now, row x and y are expressed as:







  




(2)
The output is as follows in the hidden layer:














(3)
2.2 Algorithm
1. INPUT: Intrusion attacks in 5G-enabled IoT
devices
2. OUTPUT: Accuracy of intruding attacks
3. Step 1: Collect network traffic data in a 5G-
enabled IoT environment.
4. Step 2: Pre-process the data to remove any
noise and outliers.
5. Step 3: Extract features from the data using the
Novel Convolutional Neural Network (CNN)
algorithms.
6. Step 4: Train the CNN model on the extracted
features.
7. Step 5: Use the trained model to detect any
anomalies in the network traffic.
8. Step 6: If any anomalies are detected, further
classify them as malicious or benign using deep
learning algorithms.
9. Step 7: Generate an alert for any malicious
activities detected in the network.
10. Step 8: Take the necessary action to mitigate
the attack and prevent further damage.
2.3 Deep Reinforcement Learning
Deep Reinforcement Learning (DRL) was used
(Kim) in order to develop a caching technique for the
5G and subsequent mobile networks. The findings of
their numerical data suggested that the devised DRL
caching strategy was efficient at optimising caching
resources. Moreover, this strategy was successful in
decreasing the average energy consumption of edge
Enhancing QoS in 5G IoT with CNN and Blockchain Security vs. Deep Reinforcement Learning
59
computing devices for heterogeneous 5G mobile
network technology. The Deep Reinforcement
Learning model was trained offline in a central server
with one simulated digital twin of an actual network
environment. Results showed that the system
proposed decreased the normalised energy usage
more proficiently than current approaches while
necessitating a lower computational complexity.
2.4 Algorithm
INPUT: intrusion attacks in 5G enabled IoT devices
OUTPUT: Accuracy of intruding attacks
Step 1: Initialize the Deep Reinforcement Learning
agent with appropriate hyperparameters.
Step 2: Create a state space to represent the current
environment.
Step 3: Define a reward function that rewards the
agent for successful intrusion detection and penalizes
it for false positives.
Step 4: Train the agent to recognize patterns of
malicious behavior by providing it with labeled
intrusion data.
Step 5: As the agent learns, adjust the reward function
to further refine the agent’s understanding of
malicious behavior.
Step 6: Test the agent on unseen data and measure its
performance.
Step 7 Iterate and repeat steps 4 – 6 until the agent
reaches an acceptable level of accuracy.
The aim of Deep Reinforcement Learning is to
learn an optimal policy, and maximize the discounted
total reward achieved from each state.


)

)




(4)
By taking into account the pair with the highest
Q-value, Q-learning constantly aims to select the best
course of action. In particular, DRL algorithms are
excellent at resolving issues with messaging and
mobile networks.
The proposed work was implemented using
Google Colab and TensorFlow, with a hardware
configuration comprising an Intel i5 10th generation
processor, 8GB RAM, a 1TB HDD, and Windows 10
OS. The algorithms were tested on the training sets
through empirical experiments.
2.5 Statistical Analysis
The proposed work underwent statistical analysis
using the IBM SPSS software. This, coupled with an
experimental analysis, was employed to compute the
mean and standard deviation of the dependent
variables – security, latency, accuracy, and privacy –
in relation to the independent variables: sensor ID,
sensor cycle, battery level, temperature, and time. A
lightweight consensus algorithm, the Practical
Byzantine Fault Tolerance (PBFT) (Tanwar 2021),
which eschews proof-of-work and resource-intensive
mining, was utilised to aid the implementation of the
blockchain.
3 RESULTS
The performance of the proposed CNN model was
juxtaposed against the DRL algorithm. The CNN
model reported a mean accuracy of 84.63%, whereas
the DRL algorithm achieved 70.70%. Despite the
observed variance in accuracy rates, this study
determined that there wasn't a statistically significant
difference between the two models' performance,
evidenced by a difference of 0.405 (P>0.05). The
findings indicate that the proposed CNN model
surpassed the DRL algorithm in classification
accuracy.
Table 1 details the algorithms under comparison.
The accuracy of the Deep Reinforcement Learning
network stands at 70.70%, whilst the Novel
Convolutional Neural Network's accuracy is 84.63%.
Evidently, the Convolutional Neural Network
outperforms Deep Reinforcement Learning in terms of
accuracy.
Table 2 presents statistical calculations for both the
Convolutional Neural Network and the DRL
algorithms, encompassing metrics such as mean,
standard deviation, and mean standard error. The
analysis denotes that the CNN boasts a significantly
superior mean value, 84.63, compared to DRL's 70.70.
Table 3 depicts the statistical analysis comparing
the Convolutional Neural Network and Deep
Reinforcement Learning. The latter's accuracy is
70.70%, while the Convolutional Neural Network's
accuracy peaks at 91.27%. Nonetheless, the study
discerned no statistically significant difference
between the two groups.
4 DISCUSSION
This research examined the potency of a novel
Convolutional Neural Network (CNN) architecture in
tandem with a Deep Reinforcement Learning (DRL)
algorithm and a blockchain security framework to
detect and avert intrusion attacks in an IoT-centric
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
60
Table 1: The Accuracy rate for Convolution Neural Network with N=10 sample datasets in comparison with Deep Learning
Reinforcement algorithm with the same sample size.
Sl.
Test size
CNN
DRL
1
Test size 1
76.13
67.91
2
Test size 2
76.43
68.23
3
Test size 3
77.56
68.56
4
Test size 4
82.95
69.99
5
Test size 5
85.37
69.14
6
Test size 6
87.01
69.87
7
Test size 7
88.76
70.01
8
Test size 8
90.13
72.08
9
Test size 9
90.78
75.11
10
Test size 10
91.27
76.13
Table 2: Group statistics of CNN compared with DRL algorithm in grouping with iterations of sample size 10, Mean =84.63,
standard deviation=0.70711, error mean=0.15811.
Sample (N)
Standard deviation
Standard mean error
10
6.03442
1.90825
10
2.85588
0.90311
Table 3: The independent sample tests of accuracy for CNN in comparison with DRL are kind of significantly different from
each other. There is significant difference of 0.405 (P>0.05) ensures the two groups are not significant.
accuracy
Leven’s test for
equality variance
Test for equality of means
95% of the
confidence interval
of the difference
f
sig
t
df
Sig
(2-tailed)
Mean
difference
Std. error
difference
upper
Lower
Equal
variance
assumed
8.058
0.011
6.601
18
0.405
13.9360
2.11117
18.371
9.500
Equal
variance not
assumed
6.601
12.839
0.405
13.9360
2.11117
18.502
9.369
setting. A comparative study was conducted with
extant models that utilise Deep Reinforcement
Learning to gauge if our suggested model could
amplify detection precision. Experimental findings
revealed that our model, underpinned by the CNN
architecture, attained an accuracy rate of 84.63%,
marking a substantial leap from the 70.70% secured
by the Deep Reinforcement Learning model. This
insinuates that the amalgamation of CNN, DRL, and
a blockchain security framework is proficient at
pinpointing and thwarting intrusion attacks in IoT-
driven systems.
"The Resource Allocation for Reliable Low
Latency Communication in 5G Intelligent Networks"
as posited by Tekchandani et al. (2022) substantiates
that DRL trumps primary techniques in terms of
resource consumption and drop likelihood. To bolster
the service quality in a 5G-fuelled smart grid, Qi and
Liu (2018) championed the adoption of DRL to
devise a dynamic strategy for network slice resources.
Enhancing QoS in 5G IoT with CNN and Blockchain Security vs. Deep Reinforcement Learning
61
Figure 1: Comparing the accuracy of the CNN to that of the DRL algorithm has been evaluated. The Proposed method has a
mean accuracy of 84.63 percent, whereas the DRL classification algorithm has a mean accuracy of 70.70 percent. The CNN
prediction model has a greater accuracy rate than the DRL model. This study has found that there is a statistically not
significant difference between the study groups with a difference of 0.405 (P>0.05).
This approach is nimble enough to adapt promptly to
shifts in network demand, thereby optimising
resource allocation. Li et al. (2018) put forth the
application of DQN-based 5G-V2X to refine 5G-
centric site allocation, primarily aiming to surmount
the challenge of base station allocation. To carve out
an adaptive decision-making stratagem for the initial
window in 5G MEC, Yang et al. (2021) leveraged
DQN. Their blueprint excels at boosting flow
completion while concurrently curbing congestion.
Notwithstanding the optimistic results exhibited
by the proposed model, it's imperative to
acknowledge certain inherent constraints. The
model's processing speed tends to decelerate owing to
the incorporation of maxpool layers. Moreover, CNN
models necessitate an ample volume of training data.
Though adept at detecting an array of threats, the
security framework isn't ideally streamlined for
environments with restricted bandwidth. To
circumvent these impediments, prospective
enhancements might encompass the formulation of
advanced firewall solutions to curtail data traffic and
refine the security framework to facilitate superior
bandwidth utilisation.
5 CONCLUSION
This study underscores the superior precision and
accuracy of the Novel Novel Convolutional Neural
Network (CNN) coupled with a blockchain security
framework in forecasting intrusion attacks when
juxtaposed with the Deep Reinforcement Learning
(DRL) technique. The recorded accuracy for DRL
stood at 70.70%, whereas the Novel Novel CNN
method boasted an accuracy of 84.63%. A discernible
difference of 0.405 (P>0.05) between the two
methodologies ratifies the heightened accuracy of the
Novel Novel CNN in predicting Quality of Service
(QoS) as opposed to the DRL technique. Employing
CNNs for intrusion detection offers a promising
avenue, enhancing the precision of security
frameworks. This is achieved by presenting a more
detailed portrayal of network traffic, which
subsequently augments the fidelity of intrusion
prognostications.
REFERENCES
Zhang, J.; Pan, L.; Han, Q.-L.; Chen, C.; Wen, S.; Xiang,
Y. (2021) Deep learning-based attack detection for
cyber-physical system cybersecurity: A survey.
IEEE/CAA J. Autom. Sin. 9, 377–391
Lee, I. (2020) Internet of Things (IoT) cybersecurity:
Literature review and IoT cyber risk management.
Futur. Internet, 12, 157
Almiani, M.; AbuGhazleh, A.; Al-Rahayfeh, A.; Atiewi, S.;
Razaque, A. (2020) Deep recurrent neural network for
IoT intrusion detection system. Simul. Model. Pract.
Theory, 101, 102031
Mahapatra, S., Vickram, A. S., Sridharan, T. B.,
Parameswari, R., & Pathy, M. R. (2016). Screening,
production, optimization and characterization of β-
glucosidase using microbes from shellfish waste. 3
Biotech, 6, 1-10.
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
62
Azumah, S.W.; Elsayed, N.; Adewopo, V.; Zaghloul, Z.S.;
Li, C. (2021) A deep lstm based approach for intrusion
detection iot devices network in smart home. In
Proceedings of the IEEE 7th World Forum on Internet
of Things (WF-IoT), New Orleans, LA, USA, 26–31
July 2021.
Thakkar, A.; Lohiya, R. (2021) A review on machine
learning and deep learning perspectives of IDS for IoT:
Recent updates, security issues, and challenges. Arch.
Comput. Methods Eng, 28, 3211–3243.
Li, Y.; Zuo, Y.; Song, H.; Lv, Z. Deep learning in security
of internet of things. IEEE Internet Things J. (2021);
early access. (CrossRef) 7. Idrissi, I.; Boukabous, M.;
Azizi, M.; Moussaoui, O.; El Fadili, H. Toward a deep
learning-based intrusion detection system for IoT
against botnet attacks. IAES Int. J. Artif. Intell. (IJ-AI)
2021, 10, 110.
Venkatraman, S.; Surendiran, B. (2019) Adaptive hybrid
intrusion detection system for crowd sourced
multimedia internet of things systems. Multimedia
Tools Appl, 79, 3993–4010.
Alladi, T.; Chamola, V.; Sikdar, B.; Choo, K.-K.R. (2020)
Consumer IoT: Security vulnerability case studies and
solutions. IEEE Consum. Electron. Mag, 9, 17–25
Asharf, J.; Moustafa, N.; Khurshid, H.; Debie, E.; Haider,
W.; Wahab, A. A (2020) Review of Intrusion Detection
Systems Using Machine and Deep Learning in Internet
of Things: Challenges, Solutions and Future Directions.
Electronics, 9, 1177.
Wang, X.; Zhao, Y.; Pourpanah, F. (2020) Recent advances
in deep learning. Int. J. Mach. Learn. Cybern, 11, 747–
750.
Abu Al-Haija, Q.; Zein-Sabatto, S. An efficient (2020)
deep-learning-based detection and classification system
for cyber-attacks in IoT communication networks.
Electronics, 9, 2152.
Aversano, L.; Bernardi, M.L.; Cimitile, M.; Pecori, R.
(2021) A systematic review on Deep Learning
approaches for IoT security. Comput. Sci. Rev. 2021,
40, 100389
Stefanos, T.; Lagkas, T.; Rantos, (2022) K. Deep learning
in iot intrusion detection. J. Netw. Syst. Manag. 2022,
30, 1–40.
Deena, S. R., Kumar, G., Vickram, A. S., Singhania, R. R.,
Dong, C. D., Rohini, K., ... & Ponnusamy, V. K. (2022).
Efficiency of various biofilm carriers and microbial
interactions with substrate in moving bed-biofilm
reactor for environmental wastewater treatment.
Bioresource technology, 359, 127421.
R. Pavaiyarkarasi, T. Manimegalai, S. Satheeshkumar, K.
Dhivya and G. Ramkumar, (2022)"A Productive
Feature Selection Criterion for Bot-IoT Recognition
based on Random Forest Algorithm," 2022 IEEE 11th
International Conference on Communication Systems
and Network Technologies (CSNT), Indore, India,
2022, pp. 539-545, doi:
10.1109/CSNT54456.2022.9787583.
K. Biswas and V. Muthukkumarasamy, (2016) "Securing
smart cities using blockchain technology", 2016 IEEE
18th international conference on high performance
computing and communications; IEEE 14th
international conference on smart city; IEEE 2nd
international conference on data science and systems
(HPCC/SmartCity/DSS), pp. 1392-1393, 2016.
D. Han, H. Kim and J. Jang, (2017) "Blockchain based
smart door lock system", 2017 International
Conference on Information and Communication
Technology Convergence (ICTC), pp. 1165-1167,
2017.
Dorri, S. S. Kanhere and R. Jurdak, (2017) "Towards an
optimized blockchain for iot", Proceedings of the
Second International Conference on Internet-of-Things
Design and Implementation, pp. 173-178, 2017.
Enhancing QoS in 5G IoT with CNN and Blockchain Security vs. Deep Reinforcement Learning
63