International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 543

ISSN 2229-5518

Data Security using Genetic Algorithm and Artificial

Neural Network
Mr. Mohana, K. V. K. Venugopal, Sathvik H. N.

Abstract – By making use of Genetic Algorithm, Optimization problems can be solved and the best fit individual can be
selected out of a given population. Artificial Neural Networks, a part of Artificial Intelligence, are used to simulate Human
Intelligence on a machine. Cryptography is the science and art of encrypting data so that only the intended receiver can
decrypt and retrieve the original data. This paper makes use of both Genetic Algorithm and Artificial Neural Networks for
encryption and decryption of digital data. Genetic Algorithm is firstly used to generate two random numbers, both of which
are later used as initial parameters for the Neural Network. The weights and bias of the Neural Network are created using
the initial parameters by a specific procedure as decided by the user. The receiver who knows the initial parameters can
decrypt and retrieve the original data, but the one without those parameters cannot retrieve the original data. This ensures
optimal security of data over an open channel.

Index terms – Artificial Intelligence, Artificial Neural Network, Chaotic Neural Network, Decryption, Encryption, Genetic
Algorithm,Security.

-------------------------*--------------------------

1 INTRODUCTION

IJSER
personal importance. To deal with such conflicts in
cryptography, researchers have made use of
The ability to build a secure channel is artificial intelligence, a field which uses the
one of the most challenging fields of research in concept of self-learning and adaptation by
modern communication. Since the secure channel machine itself, i.e. and intelligent or smart
has many applications, in particular for mobile machine which can change its behaviour as per the
phone, satellite and internet-based demand of the situation.
communications, there is a need for fast, effective Artificial Intelligence (AI) is a branch of
and secure transmission protocols. Nowadays, computer science that emphasis on developing
information security has become an important intelligent machines and software using applied
aspect in every organization. The people have to logic. It claims the simulation of intelligence of
be assured that the information is to be read by humans by a machine by employing reasoning,
only the sender and learning, communication and manipulation. AI is
the receiver. There comes the role of found to have intense applications in various fields
Cryptography, which ensures the secure such as medical diagnosis, stock trading, robot
transmissions of data. control, law and remote sensing. One of the
Cryptography is the exchange of important and our field in AI is Artificial Neural
confidential information among the users without Networks (ANN). An Artificial Neural Network
leakage of information to a third person. (ANN) is a mathematical model consisting of an
Cryptology was as significant as weapons during interconnected group of artificial neurons
the World War II and the Cold War. There were motivated by the working of brain. The brain
lots of studies to develop robust crypto-systems learns from experience and adaptation to
and to use them in communications. These studies environment. Also, inter neuron connection
have continued up to now. In spite of all such strengths, called weights, are used to store the
studies and implementations, this field has some acquired knowledge. ANN is a nonlinear parallel
loop holes or drawbacks. Howsoever there is adaptive system that is used to model variegated
topmost security, sometimes it happens that the relationships between inputs and outputs. The
data is leaked or stolen or manipulated by wrong output of a unit is decided by I/O characteristics
hands. Such things can be of major concerns while the overall working of ANN is determined by
sometimes, especially of national importance or its structure and the training algorithm.

IJSER © 2014
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 544
ISSN 2229-5518

Advantages of ANN include adaptive interaction encryption. Other paper present An Empirical
between elements, self-organization, real time investigation of using ANN based N-state
operation, parallel computation, and Fault sequential machine and chaotic neural networks
Tolerance. ANN helps in handling complex [3].
problems and are used in robotics, pattern Other thesis presents Genetic algorithms
recognition, medicine, manufacturing, in Cryptography [4] in which the use of genetic
optimization, signal processing, system modelling algorithms in field of cryptography is presented.
& identification, control of power-generation Other paper presents The Theory of Neural
systems. Networks and Cryptography [5] which also deals
ANN also suffers from optimization with n-state sequential machine and chaotic
problems, like getting the best values for weights. neural network. A textbook Artificial intelligence, A
One of such algorithm used for solving modern approach by J. Russell and Peter Norvig [6]
optimization problems is Genetic Algorithm ( GA ). has been referred to study some basics of Artificial
The main idea behind Genetic Algorithms (GAs) is Intelligence. [7] also presents some typical
to replicate the randomness of the nature where applications of Artificial Neural Networks in the
population of individuals adapts to its Field of Security. [8] presents Cryptography using
surroundings through natural selection process N-state Sequential Machine as well as Chaotic
and behaviour of natural system. GAs produce a Neural Network. [9] presents Chaotic Neural
population in such a way that the trait which is Generator which is used in Cryptography.
dominant, that is has higher fitness value is
replicated more likewise rest is rejected based on
threshold which is then evolved by the iterative 3 RANDOM NUMBER GENERATION
application of a set of stochastic operators like
USING GENETIC ALGORITHM

IJSER
mutation, crossover, and selection. Highest rank
signifies the better fitness.
We use both GA and ANN in cryptography Genetic Algorithm begins with an initial
to make the encryption more secure and almost population generation. Next the fitness of each
non-vulnerable to any intruder from breaking into individual is calculated according to a given fitness
it. For that, we first make use of GA to create or objective function. Based on this fitness, the
random numbers which will be serving as next generation is selected. Now two parents are
confidential parameters for initiating the chaotic selected at random from the next generation
ANN which will convert the input data bits into a individuals and reproduction is performed. For
cipher form and the original data can only be this, cross-over is done by randomly making a cut
recovered if the receiver has the information of point and converging the bits accordingly. After
both the initiating parameters. crossover, mutation is performed which involves
the flipping of any randomly selected bit in the
individual, or by using any specific mutation
function. Again a new population is generated
2 LITERATURE SURVEY which serves as the initial population for the next
generation. This process continues until the GA
Many Research papers dealing with reaches the fittest value, or it is stopped by a user
applications of Neural Networks and Genetic defined criteria. This may be the bounds of
Algorithm in cryptography have been studied and number of generations or the fitness values. After
analysed to get an idea of the previous attempts stopping the process, GA displays the final result of
made in this field. It was found that the use of the fittest individual and its fitness.
ANN along with GA has not as yet been explored. It
is essential for a key to possess randomness for
key strength and security and thus making the 4 ALGORITHM FOR CHAOTIC
code hard to break. NEURAL NETWORK
One of the papers presents Public key
cryptography using artificial Neural Networks and Artificial Neural Networks (ANN) claims
Genetic Algorithm [1] in which Random Numbers the simulation of intelligence of human by a
are generated with high randomness using genetic machine by employing reasoning, learning,
algorithm. Other paper presents cryptography communication and manipulation. We have used
using neural networks [2] in which the mutual the Chaotic Neural Network (CNN) for encryption
learning of two neural networks has been used for and decryption of data, which is different from
other neural networks in the sense that the

IJSER © 2014
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 545
ISSN 2229-5518

weights and biases of such a neural network are almost unpredictable. We can generate both the
always chaotic, i.e. extremely random and initial weight as well as the bias using GA and feed
disordered so that any outsider or third parties it to the neural network. This CNN will now
cannot break in and steal the data due to strong process the parameters according to its algorithm
encryption. The chaotic parameters for chaotic and generate encrypted keys for the given input
neural network is provided by GA. GA generate a data.
random number in every execution which is

IJSER Fig. 1. Genetic Algorithm Flowchart

Fig. 2. Chaotic Neural Network Coding Logic

IJSER © 2014
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 546
ISSN 2229-5518

5 RESULTS AND ANALYSIS for initialization and have achieved encryption of

digital data successfully. For the decrypted data to
be same as the original data, the values of initial
We have obtained the encryption of parameters must be same. This concept is
digital data in MATLAB, as mentioned in snapshot illustrated in the tables which show that different
below. And the same CNN is used to obtain the values of initial parameters will lead to different
decrypted data, provided that the weight and bias output, i.e. different from the original data.
is same as that provided in the encryption stage.
We have used the values x (0) = 0.65 and µ = 3.5

IJSER
Fig. 3. MATLAB Command Window for Encryption using CNN

Fig. 4. MATLAB Command Window for Decryption using CNN

IJSER © 2014
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 547
ISSN 2229-5518

TABLE 1
DIFFERENT INPUT DATA AND CORRESPONDING ENCRYPTED DATA

Original data Value of x(1) Value of µ Encrypted data

[1 0 1 1 0 1 1 0 1 1 0] 0.6500 3.5000 [69 111 42 131
1 129 5 130 1
129 5]
[1 0 1 0 1 1 0 0 0 1 1] 0.6500 3.5000 [69 111 42 130
0 129 4 130 0
129 4]
[0 1 1 0 0 1 1 0 1] 0.6500 3.5000 [68 110 42 130
1 129 5 130 1]
[1 1 1 1 0 0 1 0 0 1] 0.6500 3.5000 [69 110 42 131
1 128 5 130 0
129]
[1 1 0 1 1 0 0 0 0 1 1] 0.6500 3.5000 [69 110 43 131
0 128 4 130 0
129 4]

Encrypted data

[69 111 42 131 0.5500

1 129 5 130 1
IJSER TABLE 2
DECRYPTED DATA OBTAINED FROM DIFFERENT INITIAL PARAMETERS

Different values of Different values of µ

x(1)
4.5000
Decrypted data

[186 144 213

254 126 250
124
125
129 5] 254 126 5]
[69 111 42 131 0.8500 7.8000 [186 144 213 124
1 129 5 130 1 254 126 250 125
129 5] 254 126 5]
[69 111 42 131 0.7500 5.5000 [186 144 213 124
1 129 5 130 1 254 126 250 125
129 5] 254 126 5]
[69 111 42 131 0.2900 5.4500 [186 144 213 124
1 129 5 130 1 254 126 250 125
129 5] 254 126 5]
[69 111 42 131 0.6500 3.5000 [1 0 1 1 0
1 129 5 130 1 1 1 0 1 1
129 5] 0]

IJSER © 2014
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 5, Issue 2, February-2014 548
ISSN 2229-5518

6 CONCLUSION [5] I. Kanter, W. Kinzel. : “The Theory of Neural

Networks and Cryptography”, Quantum
We have Generated Random numbers Computers and Computing, V. 5, No.1, 2005
x(1) and µ using Genetic Algorithm toolbox in [6] Stuart J. Russell, Peter Norvig, “Artificial
MATLAB. The random numbers are generated Intelligence – A Modern Approach”, Prentice Hall
between the bounds 0 and 1 for x(1) and 1 and 5 2005
for µ. These two random numbers are given as
initial parameters to the Chaotic Neural Network ( [7] Navita Agarwal, Prachi Agarwal : “Use of
CNN ). CNN generates the weights and bias using a Artificial Neural Networks in the field of Security”,
specified logic or formula which are required to Vol. 3, No.1, Jan. 2013, pp. 42-44 ISSN 2230-
process the input bits. After processing, the Neural 7621 © MIT Publications
Network generates a unique key as its output for
each individual input bit given as input. In this way [8] Nitin Shukla, Abhinav Tiwari : “An Empirical
we have encrypted the original data into a form Investigation of Using ANN Based N-State
which can be reconstructed only using the two Sequential Machine and Chaotic Neural Network in
initial parameters x(1) and µ.In receiver side, after the Field of Cryptography”, Global Journal of
receiving the encrypted data now we need to Computer Science and TechnologyNeural &
decrypt it to achieve the original data. For this Artificial Intelligence Volume 12 Issue 10 Version
purpose we again need the same encryption keys 1.0 Year 2012, Online ISSN: 0975-4172 & Print
which were used as initial parameters for ISSN: 0975-4350
encryption, i.e. x(1) and µ. The algorithm for
decryption is same as the encryption, the only [9] Ilker DALKIRAN, Kenan DANIS¸MAN “Artificial
difference being the input data. Now we give the neural network based chaotic generator for

IJSER
encrypted data as input to the same CNN and get cryptology”, Turk J Elec Eng & Comp Sci, Vol.18,
the original data as the output. No.2, 2010, c_ T¨UB˙ITAK doi:10.3906/elk-0907-
140.

7 ACKNOWLEDGMENT

The authors thank and gratefully acknowledge

the support of Mr. Mohana(RVCE, Bangalore), Dr.
K. Sreelakhsmi(RVCE, Bangalore), Dr. H. V.
Kumaraswamy(RVCE, Bangalore) and Dr. B. S.
Satyanarayana(RVCE, Bangalore).

8 REFERENCES

[1] Smita Jhajharia, Swati Mishra, Siddharth Bali. :

“Public Key Cryptograpy Using Neural Networks
and Genetic Algorithms”, IEEE Trans. 978-1-4799-
0192-0/13/©2013 IEEE

[2] T. Godhavari, N. R. Alainelu and R.

Soundararajan "Cryptography using neural
network”, IEEE INDICON 2005 Conf., Chennai,
India, pp. 258-261, 11-13 Dec. 2005.

[3] https://globaljournals.org/GJCST_Volume12/3-
An-Empirical-Investigation-of Using-ANN.pdf

[4] B. Delman, “Genetic algorithms in

cryptography”, M.S. thesis, Rochester Institute of
Technology, Rochester, New York, July 2004.

IJSER © 2014
http://www.ijser.org