International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 10 Issue: 10 | Oct 2023 www.irjet.net p-ISSN: 2395-0072

Cryptanalysis of Cipher texts using Artificial Neural Networks: A

comparative study
Jaeson Karter Joseph¹, Ritu², Shresth Dewan³

¹Jaeson Karter Joseph, School of Computer Science Engineering, Vellore Institute of Technology, Vellore
²Ritu, School of Computer Science Engineering, Vellore Institute of Technology, Vellore
³Shresth Dewan, School of Computer Science Engineering, Vellore Institute of Technology, Vellore
------------------------------------------------------------------------***-------------------------------------------------------------------
Abstract - In the present times where everyone is using the old techniques, such as brute force, statistical analysis,
Internet to connect with people, socialize with them, share differential analysis, etc., in terms of speed, search space,
some information or to even find some information and efficiency in general.
regarding anything or everything, the information one needs
is available on the Internet. When everyone is using the As these techniques mature with time, we can expect lots
Internet which is available to everyone, it is a threat to an of improvement in the results that we currently get by
individual's privacy and the individual's information stored using different techniques of classification of
on the Internet. As per the report by Cybersecurity ventures cryptographic algorithms used, generations of keys, or
in 2019, the number of cyber-attacks is 158,727 per hour, finding the characteristics of the plaintext data.
2,645 per minute and 44 attacks every second all around the
world. To protect the data and information companies use Our project proposes to perform the currently available
cryptographic algorithms and security software’s. The neural networks, with varying configurations in terms of
neural networks are used in various fields like weather their designs, hidden layers, and the number of neurons in
forecasting, fraud detection, risk assessment and many due them. We look forward to performing an analysis of the
to its ability to mimic human operations and predict ciphertext to gauge the algorithm used to generate the
relationships between the vast data provided. Through ciphertext and the type of plaintext used such as whether a
Cryptography the integrity of data is ensured by using text input was used or a binary one.
hashing algorithms and message digests. It is done by
This will help us to compare the varying efficiencies of
adding codes and digital keys to ensure that what is received
different neural networks like RNN, LSTM, which maintain
is genuine and is from the intended sender, the receiver is
the varying amount of state data from the past, and CNN
assured that the data received has not been tampered with
which helps with deeper analysis of graphics, speech, and
during transmission, thus making the data reliable.
natural language.
Key Words: Cryptography, Artificial Neural 2. Literature Survey
Networks, CNN (Convolutional Neural Network), Hash
functions, RNN (Recurrent Neural Network), FC NN [1]The paper throws a light on the commonly used
(Fully connected Neural Network) techniques in analyzing the symmetric ciphers using
techniques such as boomerang technique (which does a
1. Introduction differential analysis of the cipher and the plain text when
some amount of the plain text is changed), brute force
Cryptanalysis is a highly studied field with lots of active attack (generates all possible keys and compares the
research going on to find better solutions to crack down cipher text to the plain text or the semantic analysis of the
the complex ciphers built over time such as DES, AES, generated plain text), linear cryptanalysis (to find subkeys
blowfish cipher, RSA, improve the weaknesses of ciphers in different rounds by doing a statistical analysis of the
being developed, and improve the current systems randomness trying to reduce it, by using a large number of
deployed in various organizations relating to military, cipher and plain texts.). The cryptanalysis used in the
government, public data, and confidential data system helps with the analysis of the complexity of the
incorporates. various ciphers employed by various organizations, to see
if they are well protected against attackers. It can be also
With the advent of Machine Learning approaches and Deep used to check the usage of new ciphers being developed
learning-based implementations, cryptanalysis has got a and in counteracting against illegal organizations. There
new face of exploration, with many advantages over the are some advantages to some techniques over others, like

© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 409
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 10 | Oct 2023 www.irjet.net p-ISSN: 2395-0072

some provide lesser time complexity, while some provide [5] Using the Hopfield neural network to generate binary
completeness and deterministic results. sequences and proper masking of plaintext by permutation
and calculation of generated binary sequences.
Performing cryptanalysis, requires large computation
power, and vast amounts of data in the form of plaintext - It is fast, efficient and more practical in the secure
ciphertext pairs, for getting meaningful results and still transmission of large multimedia files over public data
would be very difficult to perform on advanced ciphers communication network
like AES.
3. Overview
[2] DL based framework, and designs a DNN, using the
TensorFlow library. The attack on simple ciphers like S- Cryptography has come a long way from its beginnings in
DES, Simon and speck ciphers is performed to find the 1900 BC to current day advanced cryptography systems
text-based key used in the above ciphers. The S-DES cipher using complex mathematics and resilient implementations
was broken with a success probability of ~0.9, and ~2^8 to protect the large amounts of data being produced in
given cipher texts which reduces the search space by a today’s society. As we continue our journey in the data
factor of 8. Similarly, the Simon and speck algorithms were driven world, so does the need to combat new data
broken and 56 bits of plaintext with about ~2^12 given protection problems arise and the research for better
plaintexts. The key-space was restricted to a text-based algorithms continue.
key, which is uncommon which was done so that it is
easier to find meaningful relations between input and Cryptanalysis is one such threat to cryptography
output texts. If the key space was not limited then the DL algorithms and it is the process and study of finding
based approach failed to attack the block ciphers. vulnerabilities either in the mathematics of the algorithm
itself or the implementation of those algorithms.
[3] Feedforward BP neural network and cascaded
feedforward BP neural networks are used to carry out the With the ever-increasing amount of data, we have newer
experiment. Multiple layers (~4) of varying number of techniques to deal with the cryptanalysis of such data,
neurons are used in the BP neural network with a mean which involves using ML models to find patterns in the
square error (MSE) error function and the limit is set at ciphers, which helps in deciphering and code-breaking
0.05. Attack difficulty and data requirements are reduced encrypted messages.
in using a Neural Network based approach for
cryptanalysis. Neural networks can easily simulate the The goal of using Machine learning is to identify the type of
black box problem that cryptanalysis poses. It approaches cryptographic function used based on the input and output
the cryptanalysis problem, and finally get the algorithm parameters, finding the secret key, etc.
equivalent to the encryption and decryption algorithm.
Also, 4. Proposed System and Results

Neural networks can effectively achieve the corresponding Here we propose to find out the accuracy of correctly
Cryptanalysis results with smaller datasets. The technique predicting certain characteristics of the original plaintext
of using Artificial neural networks for cryptanalysis is using different neural networks, with varying numbers of
relatively new and doesn’t provide highly accurate results hidden layers in the network.
for complex block ciphers. Also, information regarding the
Some random data is generated using a dictionary data set,
cipher text may not always be available like the algorithm
and that is encrypted using different encryption
used, etc., and this leads to larger search space and more
algorithms. The encrypted data is then used to predict the
randomness in the estimation of the keys and plaintext.
characteristics of original data such as, the type of input
[4] Performance analysis could be done with measures like data (binary or UTF-8 format) and the cipher function
analysis of MARS and AES, Speed comparison with used.
encryption and decryption cycles, key setup and key
initialization, analysis of various key sizes, and fair speed.
Rijndael is more secure and strong when compared to the
MARS algorithm. Mars is not suitable for Smart card
implementation.

© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 410
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 10 | Oct 2023 www.irjet.net p-ISSN: 2395-0072

2. Cipher text Shift 100%

classification - FC Cipher accuracy
NN with 1 hidden was
layer achieved

3. Cipher text XOR 100 %

classification - FC Cipher accuracy
NN with 1 hidden was
layer achieved

4. Cipher classification SHIFT 100%

- FC NN with 5 and XOR accuracy
hidden layers cipher was
achieved

5. Cipher text AES 54%

classification - FC cipher accuracy
NN with 5 hidden was
layers achieved

6. Cipher text DES 70%

classification - FC cipher accuracy
NN with 5 hidden was
layers achieved

5. Conclusion
In today's world, when any information can be easily
found on the internet, it is important to protect the
information regarding any individual such as their bank
Fig -1: Flowchart of the proposed model account information, the passwords, their day-to-day
activities, etc. This is when cryptography plays a
A series of trials were carried out on cipher texts using predominant role and helps the companies to ensure their
artificial neural networks to check whether the developed customers' information is safe and not exploited by
system can predict the cryptographic algorithm used on anyone. Cryptography in the digital world is able to offer
the plain text to obtain the input cipher text, or the type of the protection for an individual’s data from an
input text. The following results were obtained during the unauthorized use of your data and fraud. Cryptography
trails: grants people the right to carry over the confidence found
in the physical world to the electronic world. It ensures
Table-1: Result Obtained people the safety to do business electronically without
worries of deception. Cryptography along with the neural
S. Test Cipher Result networks strengthens the security and builds the
No. used confidence of people about using the internet without the
fear of someone exploiting their information.
1. Plaintext N.A. 100% 6. References
classification - FC accuracy
NN with 1 hidden was 1. International Journal of Soft Computing and
layer achieved Engineering (IJSCE), ISSN: 2231-2307, Volume-3,
Issue-2, May 2013

© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 411
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 10 | Oct 2023 www.irjet.net p-ISSN: 2395-0072

2. Hindawi Security and Communication Networks

Volume 2020, Article ID 3701067,
https://doi.org/10.1155/2020/3701067

3. Hindawi Security and Communication Networks,

Volume 2018, Article ID 6868506,
https://doi.org/10.1155/2018/6868506

4. IJCSI International Journal of Computer Science

Issues, Vol. 8, Issue 4, No 1, July 2011 ISSN
(Online): 1694-0814 www.IJCSI.org

5. Chaotic neural network, received 15 March 1989,

Revised 2 October 1989, Accepted 9 January 1990,
Available online 28 August 2002, Communicated
by A.P. Fordy

© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 412