0% found this document useful (0 votes)
37 views8 pages

Prediction of Stock Market Using Artificial Neural Network

The document discusses the use of Artificial Neural Networks (ANN) for predicting stock market trends, highlighting their ability to identify complex patterns and relationships in financial data. It explains the structure and functioning of feedforward neural networks, the backpropagation training algorithm, and various prediction methods including technical and fundamental analysis. The authors emphasize the advantages of ANN in handling the complexities of stock market data and the importance of training these networks to minimize prediction errors.

Uploaded by

SHUVO 777
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
37 views8 pages

Prediction of Stock Market Using Artificial Neural Network

The document discusses the use of Artificial Neural Networks (ANN) for predicting stock market trends, highlighting their ability to identify complex patterns and relationships in financial data. It explains the structure and functioning of feedforward neural networks, the backpropagation training algorithm, and various prediction methods including technical and fundamental analysis. The authors emphasize the advantages of ANN in handling the complexities of stock market data and the importance of training these networks to minimize prediction errors.

Uploaded by

SHUVO 777
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Prediction of Stock Market Using Artificial

Neural Network
NEELIMA BUDHANI
Assistant Professor, Department of Computer Science
Amrapali Institute, Haldwani Nainital ( Uttarakhand) India.
E-mail id: neelimap28@[Link]

Dr. C. K. JHA
Associate Professor, Department of Computer Science
Banasthali Vidyapeeth, Banasthali (Rajasthan) India.
E-mail id: ckjha1@[Link]

SANDEEP K. BUDHANI
Assistant Professor, Department of Computer Science & Engineering
Graphic Era Hill University, Bhimtal Campus, Bhimtal Nainital, (Uttarakhand) India
E-mail id: sandeepbudhani13@[Link]

Abstract – Presently, all over the world


enormous amount of investment are dealing Keywords- Artificial neural network
by the Stock Markets. Nationwide financial (ANN), Stock market, prediction,
system are sturdily connected and closely backpropagation
inclined to the accomplishment of their
Stock Markets. Additionally nowadays
trading has become too reachable capital
expenditure medium, for both planned
I. Introduction
investors as well as common man also. The growth of stock market has been
identified as an economic strength of a
Artificial neural networks (ANN), belonging country. Predicting stock market return is a
to Artificial Intelligence (AI), is a technique central issue in the field of finance,
which is anticipated to identify samples engineering, mathematics, economy and
(patterns) and obtain a data model. science due to its probable finance gains [1
Significant characteristics of ANN are its 4]. Due to its ambiguous and unpredictable
potential for particular problem with nature, it takes higher hazard in comparison
progressively learning and inputs outputs to other expenditure sectors. This is one of
mapping. Neural network is well accepted the reasons which cause the stock prices
procedure to categorize unidentified, performance complex to predict. Artificial
unobserved samples in input values that is neural network (ANN) is comparatively
appropriate to predict the stock market. latest, dynamic and optimistic procedure on
Feedforward neural network with economics crisis like stock exchange index
Backpropagation training algorithm has forecast, liquidation forecasting and
been taken by us to make prediction. commercial bond categorization [2]. The

978-1-4673-9120-7/14/$31.00©2014
Authorized licensed use limited to: Dr. D. Y. Patil Educational Complex Akurdi. Downloaded on AugustIEEE
06,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
application of artificial neural networks in demonstration of artificial neurons, in which
forecast problems is awfully capable due to xi symbolizes the neuron inputs and wi
special characteristics: symbolizes the neuron weights. In general
• Artificial neural networks (ANN) are total input to the neuron is considered
general function approximation, so through a= ∑ni=0wixi. For standardization of
the relationship between the input grand total into a standard set, threshold
and productivity of the system is function (commonly used as sigmoid
discovered by it, even if this activation functions) is used and explained
relationship seems very complex. like
• Artificial neural networks (ANN),
can identify new sample even if they f (a) = 1/1+e-a …………….i
have not been in training set, this is
supposed to generalization ability of The output of this function is sure to be
neural network. within (0, 1). Fig.2 shows the threshold
• It has been verified that Multilayered function [4].
Perceptron neural network can
estimate any complex continuous
function that allow to be trained any
complex relationship connecting the
input and the productivity of the
system, this possessions of ANN is
called general function
approximation.

In 1993, a computational model for ANN


which is formed on mathematics and
algorithm are designed by Warren Fig.1 Graphical representation of artificial neuron
Mcculloch and [Link]. MP model was the
name of that model. Artificial
intelligence (AI) is the intelligence exhibited
by machines or software. ANN is a
mathematical replica imitates human mind
acquiring and decision making procedures.
Neural networks are the interrelation of
non-natural neurons running in a manner to
interpret a particular difficulty. The body of
every neuron is known as soma. A long
unevenly formed filament is dendrites and
various dendrites are linked to the soma. All
input from other neurons turn up through the
dendrites. A memory bank is used by
Fig.2 The threshold function
biological neurons to store knowledge,
though in a non-natural neuron. In the
network the signals are broadcast from input ANN generally have three layers input,
to output direction and saved as a weighted hidden and output layers. These artificial
interrelation [3]. Fig 1 expresses a pictorial neurons are prearranged within these layers.

Authorized licensed use limited to: Dr. D. Y. Patil Educational Complex Akurdi. Downloaded on August 06,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
Software interface is used to provide the architecture of this network not only holds
input values to input layer. Synapses are the the input and the output layers additionally
channel through which neurons are contain some intermediate layers. These
connected. Before addressing the input intermediate layers are also known as hidden
values toward output layer, the role of the layers.
hidden layer is to execute valuable
computations. The neurons output is Feedforward networks are usually applied as
obtained from output nodes. This completely linked network where each node
structural design is motivated by the design of one layer is associated to every node of
of brain. next layer. As a result, if P layer owns x
nodes and propels its data toward Q layer,
It is observed for some application that which contains y nodes, so each node of P
neural network has restrictions in leaning layer has y associations to calculate its
patterns or it might be volatile due to planned output, whereas each node of Q
complicated finance data applied. layer possesses x input associations.

II. Feedforward Networks


Input Hidden Output
Only two layers are there in a single layer layer layer layer
Feedforward neural network one is input and
other is output layers. Nodes at input layer
receive the input values while output is
generated by the output layer nodes. Each
node of input layer connected to all nodes of
output layers. This type of network is
acyclic in nature.

Fig.2 shows the simple Feedforward neural


network. The input layer takes input values
from software interface. Neurons are Fig.2 Feedforward Neural Network
connected to other neurons via synapses.
Input layer’s values are passed from left to Assuming x nodes in P layer and y nodes in
right, through a range of hidden layers. The Q layer let a vector symbolized by m, where
output values of neurons obtained from the mi shows threshold function output of ith
output layer. As mentioned earlier, each neurons. Let n stand for the input values for
neuron in the hidden layer has the inputs Q layer nodes, with nj be the jth neurons. Let
multiplied by a weight, and all inputs are A is matrix of x by y where Aji symbolized
summed. After this is done, the value is the association of weight from mi to nj. Take
passed to an activation function, and each into account; we proceed toward
neuron in the hidden layer then passes the single-layer Feedforward network that may
output on to neurons in an output layer, be explained as
which also multiply the values by weights
and sum them [4]. Aa = b ……….ii

Multiple layers are employed in a multilayer For multilayer Feedforward networks same
Feedforward neural network. Therefore, numerical equivalence may be used, except

Authorized licensed use limited to: Dr. D. Y. Patil Educational Complex Akurdi. Downloaded on August 06,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
A applied for every layer and when scaling primary value via investigating
proceeding to successive layers n considered associated monetary, economical and
as the value [5]. additional qualitative and quantitative issues
fundamental analysis is used. To involve
Delta Rule is the most admired learning security's assessment, containing macro
contained by a single-layer Feedforward financial aspects (similar to the generally
network, whereas Backpropagation financial system and industry circumstances)
algorithm i.e. generalization of the Delta and company-particular aspects (similar to
Rule used by multilayer Feedforward monetary condition and organization)
networks. fundamental analysts try to learn a lot. Many
performance ratios are formed like P/E ratio,
Warren Buffett. Fundamental analysis may
III. Prediction Methods propose outstanding insights, but it can be
extremely time-consuming.
The forecasting of the marketplace is with no
suspicion an out of the ordinary task.
In time series prediction mathematical
Numerous techniques are useful to
data series placed in successive manner,
accomplish the function. The above
generally takes place in equal period. In this
techniques exercise different approach,
method there is a chain of numbers
serially from natural situations (chart
composed at standard periods for a fix time
evaluation with the variation in marketplace)
duration. Time series analysis starts with
to prescribed situations (non-linear or linear
two objectives (a) recognition of characters
regressions). Categorization of these
for conditions signified by the progression
methods is below [4]:
of reviews (b) for time series variable
forecast supposed values. The above
• Technical Analysis mentioned objectives insist that sample of
• Fundamental Analysis experimental time series data is recognized
• Traditional Time Series Prediction and properly described.
• Machine Learning Methods for time series analyses separated
into two classes: frequency-domain
Technical analysis is used within finance as technique and time-domain technique.
a security analysis approach, to predict the Technique for time series data are ARIMA,
fluctuation in prices by examine past ARFIMA, GARCH, and ARMA [4].
marketplace statistics, mainly cost and
amount. This method uses charts as well as Machine learning method is
various mechanisms for recognition of remarkable for artificial intelligence
samples those guide subsequent movement. considering since it is based on the
Former behavior of shares and marketplace principle of learning from training and
signal the succeeding execution of market practice. Association models such as
by using this phenomenon the technical ANNs are well fitting for machine
analysts work. But the problem is technical learning where association weights
analysis is subjective, and our personal adjusted to improve the capability of a
biases can be reflected in the analysis. network [4].
Exponential moving average is used as
statistical techniques.

To estimate security that entails effort for

Authorized licensed use limited to: Dr. D. Y. Patil Educational Complex Akurdi. Downloaded on August 06,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
IV. Neural Network Learning backpropagation algorithm commences with
the objective to reduce error, until neural
Training is the process of learning in a neural network trained. Arbitrary weights are
network. As we require a trainer to do best in assigned at the starting of the training
any field, neural network also needs a process to normalize them as a result of
instructor that trains it to generate output. which there will be decrement in errors [6].

An error is the difference of desired output Backpropagation starts at the output layer
and actual output produced by a network. with the subsequent equations:
This error propagates in backward direction
from the output layer toward the hidden and
input layer. In the network adjustment of
weights and threshold values of neurons are
perform by this error. This process is
continuing until the network produce
minimum error. This remedial algorithm is
backpropagation algorithm, applied endlessly
and continually to inputs and consequent
outputs which are created by the reaction of
the inputs. The time taken by this process
may be long due to entire errors produced by
network go beyond an exacting rank or no
calculable errors in the network. Now it is a
trained neural network and training process
bring to an end, one can apply this neural Fig 5. Adjustment of Neuron Weights
network to generate reply for fresh input
data.

V. The Backpropagation Algorithm wij = w' ij + LR.e j . X i ……………iii

Supervised learning can be implemented by And


the backpropagation algorithm. The
backpropagation algorithm is a systematic e j = Y j .(1 − Y j ).(d j − Y j ) …………….iv
method of training multilayer ANN. It is
built on high mathematical foundation and
has very good application potential. For the ith input of the jth neuron in the output
Rumelhart, Hinton and Wilham (1986) layer, the weight wij is modifiable by totaling
applied this algorithm in multilayered neural the prior values of weights, w’ij, a
networks. According to this structure all expression regulated through the learning
neurons ordered in the form of successive rate (LR) product, an error phrase ej and
layers and spread signals in forward ways the ith input value Xi. For the jth neuron an
but the errors transmit in reverse route. error phrase ej, is calculated by the
Within the input layer, neurons accept input multiplication of the clear-cut output Yj, its
for the network, and an output layer gives negation value 1 - Yj, the variation among the
the output of the network. Hidden layers wanted output dj, and the clear-cut output.
designed by one or more middle layers. The

Authorized licensed use limited to: Dr. D. Y. Patil Educational Complex Akurdi. Downloaded on August 06,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
For the output layer the situations that create backpropagation learning.
errors are computed and adjustment of • The consequence may usually come
weights has made, the standards are traced together to any local minimum on
and the subsequent layer back is adjusted. the error plane, as stochastic gradient
The similar weight alteration process, descent is present on a plane which
determined by Equation iii, is followed; is not flat.
however the error term is produced by a • Input range or standardization
somewhat customized version of Equation required by Backpropagation
iv. learning.
• Backpropagation requires the
This modification is: activation function used by the
neurons to be differentiable.
e j = Y j .(1 − Y j ). ∑(e x .w' jx ) . ..................v
VI. WHY NEURAL NETWORK USED
The divergence between actual output and
desired output during the learning process is The prime motive to apply Neural Networks
substituted by totaling of the error phrases for in Stock Market prediction is as follows [4]:
x neurons in the layer immediately • Stock data is highly complex and
succeeding the layer being processed times hard to model, so a non-linear model
the respective pre-adjustment weights. is beneficial.
LR (learning rate) uses more or small • A large set of interacting input series
fraction of respective alteration for older is often required to explain a specific
weight. The network learn much quickly, stock, which suits Neural Network.
with the bigger value of LR but in the input
set there is a great unpredictability therefore In nature, artificial neural networks (ANN)
the network might not be trained effectively. are nonlinear like most of the natural world
Using larger rate of learning triggers bigger system. This is for the reason that the linear
adjustment in weight throughout each model is normally unsuccessful to recognize
repetition. Generally, for small values it is data pattern and evaluate at the same time as
good to put the feature with least value if the the original system is a non linear. On the
learning rate seems listless [7]. The exact other hand, a few nonlinear models like
value of learning rate depends on the common Autoregressive Conditional
application. Heteroskedasticity and Autoregressive
For numerous applications modified weigh Conditional Heteroskedasticity have been
alteration are useful which is mentioned as already applied to stock market prediction [1,
below: 4, 8].

wij = w' ij +(1 − M ).LR.e j . X j + M .( w' ij − w' ' ij ) ANN is data driven model. The uniqueness
of the neural network lies in their capability
…………vi to realize nonlinear relationship in the input
data set with no assumed declaration for the
There are a few disadvantages associated information about the connection linking the
with backpropagation learning as well: input with the output. To determine the
output values by using of input values
• The convergence is very slow for transformed by a unique function known as
backpropagation learning. activation function. They separately learn
• The convergence is not sure in

Authorized licensed use limited to: Dr. D. Y. Patil Educational Complex Akurdi. Downloaded on August 06,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
the connection from a set of labeled like [4,9]:
instruction illustration as a result in
adjustment of the network parameters [4]. • For scrutiny of credit sanction
• To judge the of risk of mortgage
Successfully Artificial Neural networks can • Prediction of financial and
be used for prediction. Some benefits of them economical issues
are automatic learning of dependencies only • Ranking the risk of investments
from precise data without any requirement to • Identification of consistencies in
add extra information. The neural network security value fluctuations.
(NN) is prepared from the past data with this • Fraud detection
anticipation that it will realize hidden
dependencies and that it will be capable to
use them for predicting into future. Artificial VII. CONCLUSION
neural network is not represented by a clearly
given model. It is further a black box that is For the growth of efficient marketing
capable to learn something. Furthermore, policies it is necessary to forecast the
when the ongoing system is dynamic and direction of changes in the stock market.
active in character, the neural network can This study attempted to predict the purpose
change its system parameters (synaptic of Artificial Neural Networks (ANN) for
weights) in actual time. Thus, neural forecasting stock market. Prediction of stock
network looks better than other models in market index is very tough task but ANN
forecasting the stock market returns [4]. has ability to predict stock index. It has been
showed that ANN is a practical, universal
Connecting one solitary result of a network approach for pattern recognition,
to a specific inside decision (identified as categorization, clustering and particularly
credit assignment problem) is very hard. for time series prediction with a great extent
Noisy data also support the negative of exactness. Nonetheless, their performance
inference of establishing wrong ancestor, for all time is not adequate.
overtraining, which will spoil generalization. Backpropagation algorithm is considered as
Lastly, a definite amount of knowledge in the best algorithm for Feedforward Neural
current subject is necessary as it is not network because it decreases errors between
unimportant to evaluate the significance of the real output and expected output in a
selected input series [4]. gradient descent method [4].
Therefore a précis of pros and cons [4]:
VIII. REFERENCES
Ï Generalization capability and
sturdiness [1] Sneha Soni, “Applications of ANN’s in Stock
Ï Mapping of input/output Market Prediction”, International Journal of
Computer Science & Engineering Technology
Ï Adjustability
(IJCSET), vol. 2, no. 3, pp. 71-83.
Ð Black-Box possessions [2] Dogac Senol, “Prediction of Stock Price Direction
Ð Overtraining by Artificial Neural Network Approach”, 2008, Ph.D
Ð Training acquires more time thesis.
[3] S. Rajasekaran, G.A. Vijyalakshi Pai. Neural
networks, Fuzzy logic and Genetic algorithms, July
In literature Neural Networks already have
2012,PHI, pp. 12-13
exercised in several commercial problems [4] Neelima Budhani, Dr. [Link] and Sandeep

Authorized licensed use limited to: Dr. D. Y. Patil Educational Complex Akurdi. Downloaded on August 06,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.
Budhani, “Application of neural network in analysis
of stock market prediction”, International Journal of
Computer Science & Engineering Technology
(IJCSET), vol. 3, April 2012, pp. 61-68.
[5] Adnan Haider and Nadeem Hanif, “Inflation
forecasting in Pakistan using Artificial Neural
Network”, Pakistan Economics and social review, vol.
47, no.1, pp 123-128
[6] Dr. Yashpal Singh and Alok Singh Chauhan,
“Neural Network in Data mining”, Journal of
Theoretical and applied information technology,
pp.37-42.
[7] Robin kumar and Amandeep Kaur Cheema,
“GPU Implementation of a Deep Learning Network
for Financial Prediction”, The International Journal
Of Science & Techno ledge (ISSN 2321 – 919X), vol.
2, issue 5
[8] Karl Nygren (2004),”Stock prediction- A Neural
Netwok Approach”. Ph.D thesis.
[9] Efstathios Kalyvas, “Using Neural networks and
Genetic algorithms to predict stock market return”,
Ph.D thesis (2001).

Authorized licensed use limited to: Dr. D. Y. Patil Educational Complex Akurdi. Downloaded on August 06,2025 at [Link] UTC from IEEE Xplore. Restrictions apply.

You might also like