A deep study of analysis for encryption and

2024 International Conference on Communication, Computing and Internet of Things (IC3IoT) | 979-8-3503-5268-9/24/$31.00 ©2024 IEEE | DOI: 10.1109/IC3IoT60841.2024.10550207

decryption algorithm in cloud data with

machine learning techniques
M. Divya Bharathi 1, B. Latha2
1Department of Computer Science and Engineering, Sri Sai Ram Engineering College, Tambaram, Chennai.
[email protected]
2Department of Computer Science and Engineering, Sri Sai Ram Engineering College, Tambaram, Chennai.

[email protected]

Abstract- An emergence of cloud computing has made secure recent approaches build upon traditional searchable
search on encrypted cloud data a popular area of study. Previous encryption methods, aiming to strike a balance between
techniques' limited ability to generate query trapdoors made them
safeguarding user access and query privacy while improving
query efficiency within the cloud environment. Current
less effective at ensuring query secrecy. A data user can also
secure query methodologies empower cloud servers to
readily examine the query results of another data user in these
efficiently search encrypted data without obtaining any
systems, as the data owner often has complete knowledge of the
valuable information regarding data files or user queries.
query results of the data users. In certain application settings, the
Nevertheless, due to constraints in the query trapdoor
data user could be reluctant to give away the privacy of their
generation process, a significant portion of these schemes,
query to anybody but themselves. It provide a search technique typically relying on symmetric encryption, can inadvertently
that enhances privacy by letting the data user create a different disclose user query privacy to certain internal entities
random query trapdoor each time and also it suggest the security external to the cloud. Data owners are depending more and
of our scheme and show through extensive experiments that it is more on different untrusted platforms (such as open cloud
exactly right. Put the suggested plan into practice and evaluate services, edges, and ML service providers) for scalability
how well it performs in terms of key generation process, secure computing or collaborative learning because of an ever-
indexing, trapdoor creation, and search timing. Compared to increasing amount of data and the necessity of creating
current hashing and attribute-based encryption searchable strong machine learning models. As a result, there is a risk
encryption technologies, the suggested scheme outperforms them. of abuse, illegal access, and privacy violations with sensitive
So, contributed to plan into practice, assess, and contrast its data and models. Through the use of cloud storage services
results using the example of searchable algorithms for encryption offered by an authorized service provider, data owners can
Keywords: Secure encrypted data, inquery result, cloud data.
escape the intricate procedures. But, there are particular
difficulties with performance, availability, security, and
scalability (PASS) in the exciting new paradigm of cloud
I INTRODUCTION
computing. According to recent study, security concerns are
the main obstacle keeping businesses and data files from
As cloud computing rapidly advances, an increasing utilizing cloud services among these obstacles.
number of organizations and individuals are embracing the
cloud to benefit from cost savings, rapid deployment, high
computation performance, and on-demand quality services.
However, the cloud, often considered a semi-trusted entity,
doesn't always inspire full trust from its users for various
reasons. Consequently, many are hesitant to entrust their
sensitive data to the cloud in its raw, unencrypted form. To
address this concern, a practical solution is to secure data
through encryption before transferring it to the cloud with
the increasing adoption of cloud computing, the secure and
efficient retrieval of encrypted data stored in the cloud has
become a significant research area. Several searchable
Fig 1.1 Basic procedure of Encryption and Decryption
encryption techniques have been created with the goal of
improving query efficiency and security, both in private key
A rise of interest in the topic has been sparked by the
and public key situations. These schemes are designed to
machine learning community's fascinating and inventive
safeguard the privacy of data access and queries for users,
research. Nowadays, there is a lot of research being done on
particularly within the context of cloud computing. Some
the issue of trustworthiness.
979-8-3503-5268-9/24/$31.00
Authorized licensed use limited©2024
to: ANNAIEEE
UNIVERSITY. Downloaded on August 22,2024 at 13:13:10 UTC from IEEE Xplore. Restrictions apply.
 II Different Encryption and Decryption
Algorithms

Machine learning and other data-driven techniques have

become essential tools in many fields of study and
application. Data owners may create complex analytical
models for a wide range of applications, including social
networking, medical informatics, entertainment, and cutting
edge research and technology, thanks to the wealth of
available data. However, due to limitations such as
inadequate internal resources, lack of experience, or need for
distributed/collaborative processing, data owners are forced
to depend on partially untrusted platforms, including cloud
Fig 2 Structure of AES and DES Algorithm.
and edge service providers, in order to handle and store large
amounts of data. As a result, cloud services related to data
B.Rivest Shamir Adleman (RSA)
analytics, such as the provision of machine learning services,
have expanded quickly in the past several years. Although all
The well-known public-key security system known as RSA
resources that are not directly held by the data owner are
was first introduced in 1978 by Rivest, Shamir, and Adleman
included in the phrase "untrusted platforms," from here on we
[11].The RSA scheme incorporates a multiplicativity
shall refer to them as "Cloud Services."
homomorphism, enabling the homomorphic encryption of
In certain applications, relying solely on training data from a
two messages modulo n. Security in RSA hinges on the
single source is insufficient in machine learning models.
computational challenge of factoring large integers, a
Therefore, many data owners wish to enable the model to
fundamental aspect of semantic security within the RSA
learn not only from their own data but also from data
framework.
originating from other sources. To accomplish this objective,
To illustrate the functionality of the homomorphic
they can utilize a cloud infrastructure, allowing the model
encryption-based RSA algorithm, consider the encryption of
hosted in the cloud to learn from their data. Although there
a rectangular area with dimensions of width 12 and height 6.
are many encryption algorithms available for encrypting data,
Utilizing the formula Cipher text = number ^ f mod (verb),
we only included those that are highly well-liked and often
with f = 79 and verb = 3337, the encryption yields cipher
utilized in professional settings including higher education,
texts of 760 and 2086. Subsequently, the area of the rectangle
the IT industry, research, and the military in this study paper.
is computed using the cipher texts: 760 * 2086 = 15,85,360.
This study discusses a number of algorithms, including
Decrypting the resulting cipher text is achieved through the
Blowfish, ARC4, MD5, AES, DES, and DES3 [12]–[14] to
method normal t = number ^ e mod (verb), with e = 1019 and
establish the correlation between the quantity of data being
verb = 3337. Decryption produces the output 72,
analyzed and the duration required for an encryption
corresponding to the original area of 12 * 6. This
technique to encrypt a given piece of data. Various sized data
demonstration elucidates how the homomorphic encryption-
files have been used for this purpose.
dependent RSA algorithm operates.
However, it's noteworthy that the key generator for this
A. Advance Encryption Standard (AES)
algorithm has limitations, and not every key generated by the
key generator supports homomorphic encryption
When the Advanced Encryption Standard, more commonly
functionality.
known as AES, is used a symmetric square code, to safeguard
group data was made by the U.S. government. Globally,
Pros: The RSA technique is frequently used for secure data
Advance Encryption Standard is utilized in programming and
transfer as it is thought to be extremely safe.
hardware to secure sensitive data, playing a crucial role in
Cons: Large key sizes are necessary for the RSA method to
cyber security, government PC security, and digital data
be safe, which entails greater processing power and storage
protection. Fig 2 describe a round key is added at the start and
capacity.
finish of encryption, followed by n rounds of application, the
last of which is unique. In order to decrypt, the opposite
procedure is used. This study examines the outcomes of
several encryption methods on files with sizes of 1KB, 10KB,
100KB, and 1MB, with encryption times expressed in
seconds.
Pros: When compared to other algorithms, AES is quicker; as
a result, it becomes much harder to hack or access the data.
Cons: It uses a lot of superfluous and sophisticated
mathematics; there are some challenges with the software's Fig 3 Structure of RSA Algorithm.
distribution and implementation.

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on August 22,2024 at 13:13:10 UTC from IEEE Xplore. Restrictions apply.
 We provide the formal definition of public-key encryption
C. Pailier Encryption Cryptosystem (PEC) (PKE) and introduce additively homomorphic public-key
encryption (PKE).
The composite Residuosity class issue forms the foundation
of the effective additive homomorphic encryption system
known as the Paillier cryptosystem (Paillier et al., 1999). This
implies that with just the encrypted messages for m1 and m2,
with the shared public key, computing the cipher text
becomes straightforward for the sum of m1 and m2. This
approach proves highly effective in safeguarding privacy,
especially in financial contexts where transactions primarily
involve addition or subtraction operations on amounts or
balances. Utilizing the public key (a, b), Paillier encryption of
a message c within the range [0, b − 1] is represented as m =
E(c, k), where k is randomly selected from [0, b − 1]. This
encryption process is probabilistic, adhering to the principles
of probabilistic public-key encryption, and possesses
homomorphic properties. Specifically, multiplying E (a1, k1)
by E(a2, k2) modulo m^2 yields an encryption of (a1 + a2)
mod n. This homomorphic feature renders the factoring-based Fig 5 Architecture of Pailier encryption
Paillier cryptosystem highly advantageous for various
cryptographic applications, particularly in scenarios involving III Comparison of Algorithms
addition or subtraction operations on encrypted values.
Decisional Composite Residuosity (DCR) implies that the
Paillier encryption method provides semantic security from Encrypti Key Bloc No of Effects
chosen-plaintext threats and is one-way private under CR. on Size(bit k rounds
Additionally, it has been demonstrated to conceal O (log n) algorith s) size
plaintext bits under a slightly stronger assumption.
ms (bits
Data privacy protection is one of the main objectives of
modern cryptography. While safeguarding private
)
information has always been important, the increasing usage AES 256 128 10,12, Efficient
of cloud services has made it even more so. Recently, 14 encryption/decry
homomorphic encryption has gained popularity for ption with
safeguarding data stored in cloud computing environments. It reduced
allows mathematical operations to be performed on a processing time.
ciphertext and return the outcome in the same manner as if RSA >1024 Min 0 Rapid encryption
the relevant plaintext had been used for the operation. 512 Round and decryption
s with a smaller
key size.
Pailier 256 128 No Save the time
rounds

Fig 6 Comparison Table

When calculations are done on data that has

been encrypted, homomorphic encryption (HE) ensures
that the outcomes are the identical as though the
calculations were to have been done with the data that
was not encrypted. In the context of the RSA algorithm,
which is one type of homomorphic encryption, this
property holds true. Here’s an example using the RSA
algorithm. Let’s say we have two plaintext numbers, 12
and 6, which are encrypted using the RSA algorithm
with the following parameters:
- Public exponent (e) = 79
- Modulus (f) = 3337
The encryption process involves raising the plaintext
Fig 4 The process of Paillier Encryption system number after which the modulus (f) with the outcome is
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on August 22,2024 at 13:13:10 UTC from IEEE Xplore. Restrictions apply.
calculated to determine the strength of the general image datasets using the HE-based RSA algorithm.
exponent (e).
So, for the number 12:
- Encrypted cipher text for 12 = \(12^{79} \mod 3337 = V CONCLUSION
760\)And for the number 6 - Encrypted cipher text for 6 The RSA algorithm is presented in this study,
= \(6^{79} \mod 3337 = 2086\) which uses homomorphic encryption to secure datasets.
To ensure the capacity to encrypt data
Now, if perform any operations on these cipher texts, homomorphically, the RSA method is used for
the result after decryption will be the same as if had encryption that is homomorphic with key generation.
performed the same operations on the plaintext. This The Spiderman relationship in ML models with data
proves the RSA encryption scheme's homomorphic protected by the HE-based RSA technique exhibits a
characteristic. The time required to use the HE-based notable improvement when compared to models created
Paillier's method and the HE-based RSA algorithm to from datasets encrypted using the Paillier's approach,
protect data and create predictive models is compared. which is based on homomorphic encryption. In
addition, the RSA method, which uses homomorphic
IV RESULT encryption, has a far shorter processing time than
Paillier's approach, which uses heuristics. As a result,
The following are the outcomes of the experiments: the HE-based RSA algorithm has been shown to is the
best method for encrypting datasets, particularly in
cases when the main goal is to use the encrypted data to
build machine learning models.

VI REFERENCES

[1] N. J. Hernandez Marcano, M. Moller, S. Hansen,

and R. H. Jacobsen authored a paper titled "On fully
homomorphic encryption for privacy-preserving deep
learning," published in the 2019 IEEE Globecom
Workshops Proceedings.

[2] The article "Towards Practical Privacy-Preserving

Decision Tree Training and Evaluation in the Cloud"
was published in the IEEE Trans. Inf. Forensics Secur.
journal in 2020. It can be accessed at doi: 10.1109/T
In Fig. 6, The Spiderman correlation between models IFS.2020.2980192. Liu, R. Chen, X. Liu, J. Su, and L.
created using samples encrypted with the RSA scheme Qiao.
and those created with Paillier's approach is
compared. [3] “Towards an efficient privacy-preserving decision
tree evaluation service in the internet of things,” L. Liu,
A comparative analysis of Spiderman relationship with J. Su, B. Zhao, Q. Wang, J. Chen, and Y. Luo,
evaluation and development of statistical models on Symmetry (Basel), vol. 12, no. 1, pp. 1–16, 2020, doi:
secured datasets RSA and Paillier's methods is shown in 10.3390/SYM12010103.
Figure 6. The Spiderman association of models created
with datasets secured using the RSA technique is shown [4] Procedia Computer Science, vol. 169, no. 2019, pp.
in orange in these visualizations, whereas the 393–399, 2020; doi:10.1016/j.procs.2020.02.235; S.
Spiderman relationship of models built with dataset Zapechnikov, "Privacy-Preserving Machine Learning as
encrypted using Paillier's approach is shown in blue. a Tool for Secure Personalized Information Services."
The present work is limited by the fact that the HE-
based RSA technique cannot be used with very big [5] "PrivEdge: From Local to Distributed Private
datasets larger than 1 GB. Furthermore, before to Training and Prediction," IEEE Trans. Inf. Forensics
encryption, categorical variables in the dataset must be Secur., vol. 15, no. April, pp. 3819–3831, 2020, doi:
removed using the HE-based RSA algorithm. All 10.1109/T IFS.2020.2988132, A. S. Shamsabadi, A.
experiments conducted utilized datasets stored in either Gascon, H. Haddadi, and A. Cavallaro.
Excel or CSV files, precluding the direct encryption of
[6] In IEEE Comput. Soc. Conf. Comput. Vis. Pattern
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on August 22,2024 at 13:13:10 UTC from IEEE Xplore. Restrictions apply.
Recognit. Work., vol. 2020-June, pp. 154–159, 2020, K.
Sarpatwar et al., "Privacy enhanced decision tree
inference," doi: 10.1109/CVPRW50498.2020.00025.

[7] A Random Decision Tree Framework for Privacy-

Preserving Data Mining, J. Vaidya, B. Shafiq, W. Fan,
D. Mehmood, and D. Lorenzi, IEEE Trans. Dependable
Secur. Comput., vol. 11, no. 5, pp. 399–411, 2014, doi:
10.1109/T DSC.2013.43.

[8] S. J. Sidi et al., "Computing with Fully

Homomorphic Encryption on Encrypted Data in the
Cloud" Data mining for the segmentation and
classification of pictures View the Cryptography project
in the volume 5, 2017, doi: 10.14738/tmlai.54.3225.
View the Computing on Encrypted Data into the Cloud
via Fully Homomorphic Encryption project.

[9] "Survey of Various Homomorphic Encryption

algorithms and Schemes," P. V. Parmar et al., 2014.

[10] In Proc. - IEEE Symp. Secur. Priv., pp. 19–38,

2017, P. Mohassel and Y. Zhang, "SecureML: A
System for Scalable Privacy-Preserving Machine
Learning," doi: 10.1109/SP.2017.12.

[11] A survey on deep learning techniques for privacy

preservation by H. C. Tanuwidjaja, R. Choi, and K.
Kim is published in Lecture Notes in Computer Science
(including subseries Lecture Notes in Artificial
Intelligence and Lecture Notes in Bioinformatics), vol.
11806 LNCS, pp. 29–46, September 2019, doi:
10.1007/978-3-030-30619-9_4.

[12] W. J. Lu and J. Sakuma, "Efficient secure matrix

multiplication for more practical privacy-preserving
machine learning as a service,"
10.1145/3267973.3267976. Proc. ACM Conf. Comput.
Commun. Secur., no. 1, pp. 25–36, 2018.

[13] "Predicting Diabetes with Machine Learning -

Pima Dataset | Kaggle."Accessed March 29, 2021 via
https://www.kaggle.com/omkarsabnis/diabetes-
prediction-using-mlpima-dataset

[14] "Standard Student Performance Dataset: UCI

Machine Learning Repository."Accessed March 29,
2021from
https://archive.ics.uci.edu/ml/datasets/Student+Perform
anc.
[15] In the Proceedings of the International Conference
on Smart Card Research and Applications, J. Daemen
and V. Rijmen, "The block cipher Rijndael,"

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on August 22,2024 at 13:13:10 UTC from IEEE Xplore. Restrictions apply.