A Multi-Criteria Intelligence AID Methodology

2023 Eighth International Conference on Science Technology Engineering and Mathematics (ICONSTEM) | 979-8-3503-4779-1/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICONSTEM56934.2023.10142810

and IoT Based Data Protection Using Machine

Learning
Dr.K.P Manikandan1, Anusha.P2, Arockia Jaya.J3, Sumit Pundir4, K.Rammohan5 and Dr.Priyabrata Adhikary6
1
Assistant Professor, Department of Computer Science Engineering, Madanapalle Institute of Technology & Science, Kadiri Road,
Angallu, Madanapalle, Andhrapradesh-517325,
2
Assistant Professor, Department of Electronics and Communication Engineering, R.M.K. Engineering College, Chennai,
TamilNadu, India
3
Associate Professor, Department of Computer Science and Engineering, Idhaya Engineering College for Women,
Kallakurichi, TamilNadu
4
Associate Professor, Department of Computer Science Engineering, Graphic Era Deemed to be University, Dehradun,
Uttrakhand, India.
5
Assistant Professor, Department of Computer Science and Engineering, St.Martin’s Engineering College, Secunderabad,
Telangana, India
6
Professor, Department of Mechanical Engineering, New Horizon College of Engineering, Bangalore, Karnataka, India

E-mail : [email protected] [email protected] [email protected] [email protected]

[email protected] [email protected]

Abstract- — IoT-powered devices have become one of become one of the key threats to people in recent
the key companions of humans in recent times. Almost times. It has been found that incidents of
every sector of the modern market is heavily reliant on cyberattacks and cyberbullying are increasing day by
IoT-powered devices such as smartphones, computers day. One of the major examples of data breaching
and other smart gadgets. This has resulted in the
has been seen during the Facebookdata privacy
massive availability of digital data containing the
personal information of the users. This has caused a scandal way back in 2018 [1]. This has made the
massive issue regarding data privacy on IoT-powered issue of data protection and privacy much more
devices in recent times. This has resulted in the serious. IoT has become one of the key technologies
application of ML-based technologies for fostering in the modern world which has billions of connected
better data protection. The concerned study has devices across the world. Progress of IoT is also
focused on the role of the different ML-powered happening at a rapid pace and it is expected to have
technologies such as federated learning and 27 billion IoT devices by 2025. This massive use
differential learning models. Thai entre study has also and growth of IoT devices have resulted in large-
focused on numerical analysis of the issues associated scale user-generated data over the internet. It has
with the mentioned models during data protection. It
been reported that the amount of user-generated data
has also provided survey data regarding different FL-
based algorithms such as LM, NN, NM, DT and CM. will stand between 4 zettabytes to 140 zettabytes
This study has analyzed the issues and strengths of between2020-2025 [2]. A significant amount of
ML-based data protection on local and global data personal data of users is collected through IoT
sets. devices which can make the user's privacy very
vulnerable. It has resulted in reviewing the privacy
Keywords: federated learning, differential learning, machine policies and strategies associated with digital data
learning, distributed learning, cryptographic methods. systems. Concerning this, the evolution of new
technologies like machine learning has become
I.INTRODUCTION very crucial in recent times. Machine learning can be
one such technology which can help in fostering the
Technology has become one of the key factors in the data protection regime in IoT-powered devices and
modern-day world with its presence in almost every systems. However, ML can also play some different
sector. People are surrounded by technology at every roles to create problems regarding privacy protection
step and entire daily life operations are becoming in IoT-powered devices. This has made this entire
digital day by day. This has made a huge presence of study identify the opportunities of ML for data
digital data of people on internet media. This has

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 12,2023 at 07:31:31 UTC from IEEE Xplore. Restrictions apply.
protection regimes in IoT devices. However, besides
that, the entire study has also highlighted the
problems and privacy risks with different ML
models and systems. This entire study has attempted
to provide a comprehensive analysis and
investigation of the efficiency of ML in data
protection through different issues related to IoT-
powered tools. This entire study has also come up
with a detailed methodology for efficient data
collection regarding the mentioned issue. This entire
study has come up with numerical formulas for
analyzing the issues and problems in the ML and
FL-based data protection system. It has provided
survey data based on different algorithms used in the
SGD model of the FL learning system which is one
of the latest advancements in traditional ML
techniques. This analysis has helped the study to
also formulate future steps for better data protection
in future. Figure 1: IoT Data Protection (Source: [3])

II.METHODOLOGY One such technique is Differential privacy which is

widely used for data protection for IoT-powered
Research methodology is one of the most important tools. This type of data protection technique adds
segments for a study to attain its proposed outcomes. data perturbation with the original user-generated
The selection of the right methodology allows a data for executing data protection [3]. It has also
study to execute better data collection as well as collected secondary survey data regarding
better analysis of the collected data. This entire distributed learning. The concerned study has used
study has mainly focused on secondary data a quantitative approach. This approach has helped
collection methods. The concerned study has the concerned study to execute mathematical
collected both qualitative and quantitative secondary analysis of the collected secondary data which has
data from authentic sources. It has focused on survey resulted in better outcomes for the study.
data for identifying the impact of ML in protecting
the data privacy of users. The concerned study has Numerical analysis of machine learning model for
collected survey data for identifying the efficiency data protection
of ML in providing data safety over mobile
networks. This survey data has also been used for 𝜃 ∗ = arg min⁡
( 𝐿 𝑦𝑖, 𝑓𝜃 𝑥𝑖 + Ω(θ)),
analyzing the impact of ML on data protection in 𝑖
cloud computing. This entire study has also
The entire study has also collected numerical data
collected secondary data through surveys for
regarding different ML-based data protection
identifying the issues and threats regarding data
models. It can be found from the above formula that
privacy in IoT. The concerned study has focused on
collecting data regarding “centralized ML-based (𝑓𝜃) is denoting the imputed data which is the x
privacy protection solutions” which can be very vector concerning the output data which is denoted
effective for providing future steps in attaining better by the y vector[4].Y is the set of different data
results regarding data protection. Data regarding classes and x^d is the vector space of d-
encryption techniques and access control in dimensional data sets. This numerical formula has
centralized encryption has also been collected in been developed for the training model of ML-based
this study. The collection of this type of data has data protection. This model can be very efficient for
helped the study to perform a better analysis of finding the optimal parameter regarding the accurate
results regarding the concerned topic. Different relationship between the input and output data.
techniques and models have been identified during
the data collection method. Data fusion, model building and data processing
are the three key steps associated with the

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 12,2023 at 07:31:31 UTC from IEEE Xplore. Restrictions apply.
centralized learning method. For executing
centralized learning methods, this study collected
numerical data regarding federated learning and
gated recruiting units. The FL-based data protection
reads data sets which are stored in single or globally
predicted models. This has been one of the key
problems in FL-based data protection.

Therefore, Numerical formula for the globally

predicted model problem of the data set

The above formula is describing the global model of

data aggregation. In the above formula, DK is the
local storage of data sets in the IoT-powered devices
K. The above formula is used for identifying the
upgrade in the data sets which can be very helpful
for fostering better data protection in future [5]. The
main purpose of the mentioned numerical formula in
figure 2 is to identify the output data with the loss Figure 2: Survey regarding ML-based privacy
function. This can be very effective for identifying protection (Source [6])
the issues with ML and FL-powered data protection
systems in future. In the above formulas, f(ω) is It can be analyzed from the above survey data that
denoting the data set dimension in which the input an ML-based data protection system has efficiency
data is denoted concerning output data. over both mobile networks and cloud computing.
The above data has helped to analyze that ML-based
data protection system provides solutions for three
III.RESULT AND DISCUSSION
different areas such as malware, privacy and
network. The above survey data has also collected
Machine Learning based cyber security
data regarding the Federated Learning model of
Machine learning has been one of the key AI-based
data protection. The idea of federated learning was
technologies which help to attain better data
introduced first by Google in 2017. One of the key
protection. ML-based data protection technology
features of FL has been that it allows data scientists
helps to get a quick insight into threats regarding
to link and train statistical models of decentralized
digital data protection. The survey data regarding
devices with the local data set [7]. This has omitted
different models of AI-powered data protection is
the necessity of uploading any private data to the
given in the below images.
cloud storage for training purposes of the data
scientists. This has been one of the potential
differences between ML-based data privacy and FL-
based data privacy. It has been seen that in
traditional ML-based data protection systems data
sets are required to be uploaded on a single server.
This has allowed FL-based data protection systems
to provide better privacy protection which is also
clear from the survey data. Provide better data
protection compared to traditional ML-based data
protection.

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 12,2023 at 07:31:31 UTC from IEEE Xplore. Restrictions apply.
 Table 1: Comparison between different ML-based
models for data protection

The above table is showing a detailed comparison

between different models of ML-based data
protection. As mentioned earlier FL-based data (Source: [8])
protection is one such ML model of data protection
which does not require any new data sets and this However, the mentioned algorithms can be very very
feature has made it much more reliable for data problematic during the upgradation of local data set
protection in future. The above table is showcasing models as well as the global models. These issues
that FL-based data protection is based on SGD [9]. can be addressed by framing a personalized
The above table is also showcasing different SGD- federated learning algorithm [11]. This can be very
based algorithms such as LL, NM and DT which helpful for allowing the data scientist to read the
are being used for showcasing neural networks and personalized models of the local data set which can
linear models of data privacy in IoT-powered tools result in better data privacy in IoT devices in future.
and devices. In the above table, CM has been used
for indicating cryptographic methods and DP has IV.CONCLUSION AND FUTURE DIRECTION
been used for indicating differential privacy during
data protection in IoT devices [10]. It can be stated This entire study has identified different models and
by analyzing the above table that the algorithms algorithms which can be used for ML-based data
named LM, and NN are efficient for centralized protection in IoT devices. It can be concluded from
learning models during ML and FL-based data the study that traditional ML-based techniques can
protection. It has been also analyzed from the above be better replaced with FL-based models. This
table that algorithms like CM and DP are efficient advanced ML powered model can provide data
for both the centralized and decentralized data privacy without asking for a new data set or user
protection regime in IoT-powered devices. generated information. However, through the
mathematical formula it has been identified that
SGD and FL-based models have some issues
regarding up gradation of global data sets. This can

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 12,2023 at 07:31:31 UTC from IEEE Xplore. Restrictions apply.
be improved in future with more advanced ML [10] Chopra, P., Gollamandala, V.S., Ahmed, A.N., Babu, S.T.,
Kaur, C., Prasad, N.A. and Nuagah, S.J., 2022. Research Article
algorithms. This entire study has also provided a list Automated Registration of Multiangle SAR Images Using
of algorithms which can help in advance of Artificial Intelligence.
[11] Babu, S.T. and Rao, C.S., 2020, July. Statistical features based
cryptographic methods for data protection. This optimized technique for copy move forgery detection. In 2020
study may also help in betterment of differential 11th International Conference on Computing, Communication
learning method through CM and DP algorithms for and Networking Technologies (ICCCNT) (pp. 1-6). IEEE.
[12] Kalimuthukumar Sakthivel, Rajesh Krishnasamy, Kannapiran
data protection in future. Balasubramanian, Vijayakumar Krishnakumar , Manikandan
Ganesan “ A revolutionary Partial Resonant Inverter and
REFERENCES doubler rectifier with MPPT based on Sliding Mode Controller
for harvesting Solar photovoltaic sources”, Sustainable
Computing: Informatics and Systems, Volume 36 December
[1] Liu, B., Ding, M., Shaham, S., Rahayu, W., Farokhi, F. and Lin,
2022 100811. https://doi.org/10.1016/j.suscom.2022.100811
Z., 2021. When machine learning meets privacy: A survey and
[13] Kalimuthukumar Sakthivl,Rajesh Krishnasamy, Kannapiran
outlook. ACM Computing Surveys (CSUR), 54(2), pp.1-36.
Balasubramanian, Vijayakumar Krishnakumar and Manikandan
[2] Rodríguez, E., Otero, B. and Canal, R., 2023. A Survey of
Ganesan. “Averaged state space modelling and the applicability
Machine and Deep Learning Methods for Privacy Protection in
of the series Compensated Buck-Boost converter for harvesting
the Internet of Things. Sensors, 23(3), p.1252.
solar Photo Voltaic energy”, Sustainable Energy Technologies
[3] Zhao, P., Zhang, G., Wan, S., Liu, G. and Umer, T., 2020. A
and Assessments, Volume 53, Part C, October 2022, 102611
survey of local differential privacy for securing internet of
[14] T. Vino,S.S.Sivaraju,R. V.V. Krishna,T. Karthikeyan,Yogesh
vehicles. The Journal of Supercomputing, 76, pp.8391-8412.
kumar Sharma, K.G.S.Venkatesan, G.Manikandan,
[4] Liu, B., Ding, M., Shaham, S., Rahayu, W., Farokhi, F. and Lin,
R.Selvameena,and Mebratu Markos “Multicluster Analysis and
Z., 2021. When machine learning meets privacy: A survey and
Design of Hybrid Wireless Sensor Networks Using Solar
outlook. ACM Computing Surveys (CSUR), 54(2), pp.1-36.
Energy”, Hindawi, International Journal of Photoenergy,
[5] Liu, Y., James, J.Q., Kang, J., Niyato, D. and Zhang, S., 2020.
Volume 2022, Article ID 1164613, 8 pages,
Privacy-preserving traffic flow prediction: A federated learning
https://doi.org/10.1155/2022/1164613.
approach. IEEE Internet of Things Journal, 7(8), pp.7751-7763.
[15] Manikandan, G. and Anand, M. “SEC-TAED Based Error
[6] Rodríguez, E., Otero, B. and Canal, R., 2023. A Survey of
Detection and Correction technique for data transmission
Machine and Deep Learning Methods for Privacy Protection in
systems”, Indonesian Journal of Electrical Engineering and
the Internet of Things. Sensors, 23(3), p.1252.
Computer Science, Vol.10,No.2, May 2018, pp.696-703,
[7] ODSC, 2023. ““How You Can Use Federated Learning for
ISSN:2502-4752
Security & Privacy". Available at:
[16] S B G Tilak Babu and Ch Srinivasa Rao, "An optimized
https://odsc.medium.com/how-you-can-use-federated-learning-
technique for copy–move forgery localization using statistical
for-security-privacy-ee0c99cf54b3 [Accessed on 14/03/2023
features", ICT Express, Volume 8, Issue 2, Pages 244-249,
[8] Li, Q., Wen, Z., Wu, Z., Hu, S., Wang, N., Li, Y., Liu, X. and
2022.
He, B., 2021. A survey on federated learning systems: vision,
[17] Ashokkumar, N., and A. Kavitha. "Transition level energy
hype and reality for data privacy and protection. IEEE
consumption of NoC (network-on-chip) using data encoding
Transactions on Knowledge and Data Engineering.
techniques." 2015 2nd International Conference on Computing
[9] Chopra, P., Gollamandala, V.S., Ahmed, A.N., Babu, S.B.G.,
for Sustainable Global Development (INDIACom). IEEE, 2015.
Kaur, C., Achyutha Prasad, N. and Nuagah, S.J., 2022.
Automated Registration of Multiangle SAR Images Using
Artificial Intelligence. Mobile Information Systems, 2022.

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on June 12,2023 at 07:31:31 UTC from IEEE Xplore. Restrictions apply.