Volume: 02, December 2013, Pages: 440-449
International Journal of Computing Algorithm
Forecasting of Stock Prices Using Multi Layer Perceptron
A. Victor Devadoss1, T. Antony Alphonnse Ligori2
1
Head and Associate Professor, Department of Mathematics, Loyola College, Chennai, India.
2
Ph. D Research Scholar, Department of Mathematics,Loyola College, Chennai, India.
[email protected],
[email protected]
Abstract
Prediction of stock market has been a challenging task and of great interest for researchers
as the very fact that stock market is a highly volatile in its behavior. For predicting stock
price of Bombay Stock Exchange (BSE), Multilayer Networks with dynamic back propagation
has been used. The stock prices are determined and compared with two different
architectures NN1 (3-16-1) and NN2 (3-6-1). Neural Network based forecasting of stock
prices of selected sectors under Bombay Stock Exchange show that neural networks have the
power to predict prices albeit the volatility in the markets. The paper is organized as follows.
In Section one the volatile nature of stock market is discussed. Section two reviews the
literature on the applications of ANNs in predicting the stock prices. Section three gives an
overview of forecasting methods. In Section four the concept of Artificial Neural Network
presented. Section five presents the methodology adopted in forecasting the stock price. In
the final section results, future direction of the study and conclusion are derived.
Keywords -Artificial Neural Network (ANN); Fundamental Analysis; Stock Market;
Technical Analysis; Time Series Analysis.
I.
INTRODUCTION
Prediction of stock market is substantial in
finance and is gaining more attention, due
to the fact that if the direction of the
market is predicted successfully the
investors may be better guided.
Researchers have proposed many models
using various fundamental, technical and
time series forecasting techniques to give
competitive predictions. The existence of
the nonlinearity and volatility of the
financial market is propounded by many
researchers and financial analysts [1]. As
the stock market being dynamic, nonlinear, and chaotic in nature it is very
difficult to understand because of its
volatility, hence it is of great importance
for the investors to know its behavior
which would help for their effective
investment in it.
Artificial Neural
Network (ANN) has the ability to discover
the nonlinear relationship in the input data
set without a priori assumption of
knowledge of relation between the input
and the output [2]. Hence Artificial Neural
Integrated Intelligent Research (IIR)
Networks suits well than other models in
predicting the stock market returns.
Nowadays, stock markets have become an
integral part of the global economy. Any
fluctuation in the market influences our
personal and corporate financial lives and
the economic health of a country. Due to
its unpredictable behavior there is always
some risk to the investment in the stock
market [3]. Many scientific attempts have
been conducted on stock market to extract
some useful patterns and predict their
movements. Moreover, many financial
corporations have been in trying to model
the behavior of stock movements since it
can guide their financial investments and
yield significant profits. Albeit no method
has been successful to accurately predict
the stock price movement till now. ANNs
ascertain a hope for the investors, an
ability to predict the stock prices.
II. LITERATURE REVIEW
Artificial Neural Networks are highly
flexible function approximators that can
map any nonlinear function [4] and were
440
Volume: 02, December 2013, Pages: 440-449
International Journal of Computing Algorithm
used initially in the fields of cognitive
science and engineering and later applied
in financial time series forecasting. ANNs
are also being used for a wide variety of
tasks in many different fields of business,
industry and science.
One major
application area of ANNs is forecasting
[5]. ANNs are well suited for problems
whose solutions require knowledge that is
difficult to specify but for which there are
enough data or observations [6]. The first
forecasting using ANNs [6] using
Wildrow’s adaptive linear network to
weather forecasting. The first study to use
the multi-layer feedforward networks for
forecasting purposes is seen in the study
[7]. A comprehensive study is done by
researchers on the common parameters in
designing a backpropagation neural
network and provided step by step
methodology to design a neural network
for forecasting economic time series data
[8].
later selling them when they become overpriced. Fundamental analysis is more
useful for long-term investors.
B. Technical analysis
The technical analysis is characterized by
a large number of rules and indicators
committed to identify and explain the
regularity of historical price dynamics.
Technical analysis uses patterns of the
price history of a financial instrument in
order to provide indications on the future
behavior of prices [9]. Technical analysts
argue that prices gradually adjust to new
information.
The Moving Average
method (MA) is one of the most used
methods of technical analysis.
This
method involves a comparison of the
market prices or index with the long MA.
The MA method is easy to use and apply
in
investment
decision-making
or
empirical tests [10]. The research [11]
showed that MA method can generate
significant forecast value errors and
deviations from real prices and is not
successful in price movement trend
generation.
Technical analysis is commonly used for
taking ‘buying’ and ‘selling’ decisions in
the stock market. This analysis attempts to
predict the future price of a particular
share on the basis of a study of its price
movements in the past. Technical analysts
are also called as ‘chartists’ as they use
charts and graphs for keeping a record of
share price movements. They believe that
an elaborate study of share price charts and
graphs will reveal regular and recurrent
patterns of price behavior which are likely
to be repeated in the future. Technical
variables most frequently cited are shown
[12]. They usually ignore all fundamental
data like sales, earnings, profits, dividends,
business prospects of the company, etc.
and believe that these factors have already
been taken into account by the market and
are fully reflected in the current market
price of a share. Technical analysis by the
very nature of its approach is suitable for
speculators and short-term traders in
shares.
III. METHODS USED FOR FORECASTING
Let us enumerate some available
forecasting methods in predicting the stock
prices.
A. Fundamental analysis
Fundamental analysis is a type of
investment analysis adopted by investors
for taking investment decisions and the
investors who follow this approach are
called ‘fundamentalists’. They try to
estimate the intrinsic worth of a
company’s share, by studying its sales,
earnings, profits, dividends, management
proficiency, and a host of other economic
factors that have a bearing on the
company’s profitability and business
prospects. The objective is to estimate
what the price of a particular company’s
share out to be and consider this price to
be its intrinsic or true value of the share as
it reflects the inherent worth and value.
With the help of intrinsic price one can
judge whether the shares are currently
over-priced or underpriced in the stock
market. The fundamentalist makes his
money by buying underpriced shares and
Integrated Intelligent Research (IIR)
441
Volume: 02, December 2013, Pages: 440-449
International Journal of Computing Algorithm
C. Time series forecasting
pattern. The power of neural networks
comes to life when a pattern that has no
output associated with it, is given as an
input. In this case, the network gives the
output that corresponds to a taught input
pattern that is least different from the
given pattern.
The study of artificial neural networks has
been inspired by the biological learning
systems which consist of very complex
webs of interconnected neurons. ANNs
are built out of densely interconnected
units (neurons) where each unit takes a
number of real-valued inputs which
produces a single real-valued output that
may in turn be an input to other units.
ANNs have the ability to learn and thereby
acquire knowledge and make it available
for use. ANNs are among the most
effective learning methods to learn and
interpret complex real-world sensor data
[14]. We just recall the notion of neural
network called the Weighted Multi Expert
Neural
Network
(Wt.M.E.N.N)
constructed using the fuzzy neural
networks. This Wt.M.E.N.N., guarantees
equal representation of opinion of each
expert; hence this method has an
advantage over the Fuzzy Neural
Networks. Neural Network learning can be
either supervised one or an unsupervised
one. In a supervised learning algorithm,
learning is guided by specifying, for each
training input pattern the class to which the
pattern is supposed to belong. In an
unsupervised one, the network forms its
own classification of patterns. The
classification is based on commonalties in
certain features of input pattern. Since the
data is an unsupervised one, we make use
of Wt.M.E.N.N. In any supervised
learning, a training set of correct inputoutput pairs is given so as to minimize the
error, but in an unsupervised one the
output is purely based on the input data.
We just recall the definition of Neural
Network.
Definition 1
A neural network is a computational
structure that is inspired by observed
Time series forecasting is the analysis of
the time series data that tries to predict the
near future data based on its past data.
This is significant in the field of stock
market investment, as investors want to
make right decisions at right times to
maximize
their
financial
profit.
Conventional researches used time series
analysis techniques like mixed auto
regression moving average (ARMA) and
multiple regression models [13]. Time
series forecasting usually find a trend in
the past data to predict future data. The
more past data, the easier it is to find a
pattern. However if the history of a stock
is short, an accurate analysis and forecast
for such little past data is difficult. So, in
this case neural networks are described as
great tools to use in this scenario.
D. Artificial Neural Networks in stock
market prediction
Investment in stock market carries a higher
risk due to its uncertainty and volatility
and hence forecasting the stock price
behavior is very difficult. The difficulty
arises due to the nonlinear and complex
behavior of stock prices. As the primary
application of artificial neural networks is
in areas where problems are ill-defined,
data is incomplete or noisy in nature and
the environment itself is dynamic. As
artificial neural networks are able to adapt
to noisy data and establish input-output
relationship of nonlinear data, the behavior
of stock price prediction is possible. In the
last two decades extensive researches been
attempted through neural networks to
forecast stock prices.
ARTIFICIAL NEURAL NETWORKS
An important application of neural
networks is pattern recognition. Pattern
recognition can be implemented by using a
feed-forward neural network that has been
trained accordingly. During training, the
network is trained to associate outputs
with input patterns. When the network is
used, it identifies the input pattern and
tries to output the associated output
IV.
Integrated Intelligent Research (IIR)
442
Volume: 02, December 2013, Pages: 440-449
International Journal of Computing Algorithm
process in natural network of biological
neurons in the brain. It consists of simple
computational units, called neurons that are
highly
interconnected.
Each
interconnection has a strength that is
expressed by a number referred as weight.
Definition 2
The bias defines the value of the weighted
sum of inputs around which the output of
neuron is most sensitive to changes in the
sum.
Now we proceed on to define the notion of
Weighted Multi Expert Neural Network.
In Neural Network bias plays an important
role. So we take the bias as an input with
value -1 and its corresponding weight is
the sum of the average of the other input
weights.
In general, using this newly constructed
Weighted Multi Expert Neural Network
(Wt. M.E.N.N.), we can extend to `n’
number of experts.
The class of sigmoid function S,
defined by the formula
S β (a) = (1 + exp {-βa})-1 .
Then, the output of the neuron is defined
architecture of an ANN, these methods are
usually quite complex in nature and are
difficult to implement. Furthermore none
of these methods can guarantee the
optimal solution for all real forecasting
problems. To date, there is no simple
clear-cut method for determination of
these parameters. Guidelines are either
heuristic or based on simulations derived
from limited experiments. Hence the
design of an ANN is more of an art than a
science.
An artificial neural network is defined as a
data processing system consisting of a
large
number
of
simple
highly
interconnected
processing
artificial
neurons in an architecture inspired by the
structure of the cerebral cortex of the
brain. There are several classes of neural
networks. It is classified according to the
learning mechanisms. The three broadly
classified learning methods are supervised
learning, unsupervised learning and
reinforced learning. There are three
fundamental classes of networks namely,
single layer Feedforward network,
multilayer Feedforward network and
recurrent network.
B. Multilayer feedforward network with
back propagation algorithm
MLP is a feedforward neural network with
one or more layers between input and
output layer. Feedforward means that data
flows in one direction from input to output
layer. MLP has three layers; an input
layer, one more hidden layers and output
layer. The input data are fed to the
neurons in the input layer and after
processing within the individual neurons
of the input layer the output values are
forwarded to neurons in the hidden layer
and finally to the neurons in the output
layer. MLPs are widely used for pattern
classification, recognition, prediction and
approximation.
Connections among the neurons are
associated by weights and changing the
weights in a specific manner results to
learning of the associated network. The
procedure by which the weight changes
by
n
Y S Wi X i -
i 1
where is a positive constant (Steepness
parameter), is called the bias of the
neuron, since the bias is considered as an
input, x0 = –1 and the associated weight
w0 = . The quantities Xi and Wi denote the
inputs and weights respectively.
A. Neural Network Architectures
An ANN is typically composed of layers
of nodes. In the popular MLP, all the
input nodes are in one input layer, all the
output nodes are distributed into one or
more hidden layers in between. An MLP
is determined by the following variables:
The number of input nodes
The number of hidden layers and
hidden nodes
The number of output nodes.
The selection of these parameters is
basically problem-dependent. There exists
many different approaches such as the
pruning algorithm for finding the optimal
Integrated Intelligent Research (IIR)
443
Volume: 02, December 2013, Pages: 440-449
International Journal of Computing Algorithm
take place in the network is called learning
or training algorithm. The backpropagation
algorithm is the most commonly used
learning technique. The technique consists
of a forward pass and a backward pass. In
the forward pass, an input vector is applied
to the nodes of the network and result of
which becomes a set of outputs for the
network at the output layer. During this
phase the weights are all fixed. In the
backward pass, the error term is calculated
by finding the difference between actual
response of the network and desired
response specified to the network and is
propagated backward through the network.
Here the weights are adjusted so as to
make the actual response of the network
becomes closer to desired response [15].
The neural network training is an
unconstrained nonlinear minimization
problem in which synaptic weights of a
network are iteratively modified to
minimize the overall mean or total squared
error between the desired and actual output
values.
The most popularly used
backpropagation algorithm is used for
training which follows the gradient
steepest descent method. For the gradient
descent algorithm, a step size, called
learning rate must be specified.
The learning rate is a constant of
proportionality which determines the size
of the weight changes. The weight change
of a neuron is proportional to the impact of
the weight from that neuron on the error. A
very small learning rate requires more
training time. One method to increase the
learning rate and thereby speed up training
time without leading to oscillation is to
include a momentum term in the
backpropagation learning rule.
The
momentum term determines how past
weight changes affect current weight
changes. Most neural network software
programs provide default values for
learning rate and momentum that typically
work well. Initial learning rates in the
literature are found to vary widely from
0.1 to 0.9. Common practice is to start
training with a higher learning rate such as
0.7 and decrease as training proceeds.
Many neural network programs will
automatically decrease the learning rate
and increase momentum values as
convergence is reached.
Integrated Intelligent Research (IIR)
V. APPLICATION OF MLP IN
FORECASTING STOCK PRICE
Assume that xi is the data series of
stock price, where i 1,2,3,..., N and N is
the number of data of selected companies
under Bombay Stock Exchange between
1stJanuary 2012and 7thNovember 2013.
The following Table I display the selected
companies and sectors used in the study.
TABLE I. SELECTED COMPANIES
UNDER BSE FOR THE PRESENT
STUDY
Sl
Sector
Company
No.
Tata
Computers1
Consultancy
Software
Services Ltd
Infosys
Computers2
Technologies
Software
Ltd
Dr. Reddy’s
Healthcare and
Laboratories
3
Pharmaceuticals
Ltd
Sun
Healthcare and
4
Pharmaceutical
Pharmaceuticals
Ltd
In the present
study Multilayer
feedforward network (MLP) with dynamic
backpropagation learning has been used.
Two networks are constructed namely
NN1 and NN2. Both the network contains
an input layer, one hidden layer and one
output layer. The number of neurons in
the input layer is three which are the three
consecutive past prices of the particular
stock under study. The number of hidden
layers and the number of neurons selected
in the study has been done heuristically.
The hidden layer consists of sixteen
neurons in NN1 and six in NN2 which
provide the network with its ability to
444
Volume: 02, December 2013, Pages: 440-449
International Journal of Computing Algorithm
generalize. In the study it was started with
a fewer number of neurons, but due to the
inaccuracy in the initial predictions the
number of neurons were increased in the
hidden layer [16].
The number of neurons in the output layer
is one as the modeling applied in the study
aims to predict one step ahead closing
value in the future forecasting.
The
sigmoid function is used as a transfer
function because it is commonly used for
time series data as they are nonlinear and
continuously differentiable which are
desirable properties for network training.
Table II and Table III show the parameters
for the construction of training sets for the
companies. In the following tables LR
denotes learning rate and MR denote
momentum rate.
TABLE II. Training Parameters for NN1
To measure the performance of the neural
network model used, Mean Absolute
Percentage Error (MAPE), Mean Absolute
Deviation (MAD) and Root Mean Squared
Error (RMSE) were calculated. Suppose
(a1, a2 ,a3 ,..., an ) are
actual values and
(p1 ,p2 ,p3 ,...,pn ) are the predicted values then
the MAPE, MAD and RMSE can be
calculated by using the formula (1), (2)
and (3).
Company
Name
Tata
Consultancy
Services Ltd
Infosys
Technologies
Ltd
Dr. Reddy’s
Laboratories
Ltd
Sun
Pharmaceutical
Ltd
Total
Mean
Squared
Error
LR
MR
Total
Net
Error
0.4
0.5
0.00035
0.00101
0.4
0.4
0.00117
0.00382
0.2
0.5
0.00036
0.00069
0.2
0.3
0.00030
0.00114
TABLE III. TRAINING PARAMETERS
FOR NN2
Company
Name
LR
MR
Total
Net
Error
Tata
Consultancy
Services Ltd
Infosys
Technologies
Ltd
Dr. Reddy’s
Laboratories
Ltd
Sun
Pharmaceutical
Ltd
0.5
0.5
0.00032
Total
Mean
Squared
Error
0.00045
0.4
0.3
0.00125
0.00330
0.7
0.2
0.00062
0.01033
0.7
0.5
0.00018
0.00187
Integrated Intelligent Research (IIR)
MAPE 100 *
MAD
1
a p
ia i
n
i
a p
i
n
i
RMSE mean( ai pi ) 2
--- (1)
--- (2)
--- (3)
In this study NEUROPH has been
employed to forecast stock prices of
selected companies and sectors listed in
the table1.
Step1: Divide the whole data into two
teams 60% for training set and 40% for
testing set.
Step 2: 60% data has been used to train
the network. The normalization for the
input
is
done
using
the
x xmin
, where x denotes
formula x N
xmax xmin
the value that should be normalized; xN
denotes the normalized value of x; xmin
represents the minimum value of x;
xmax represents the maximum value of x.
After the normalization the data (stock
prices) will be in the range of [0, 1].
Step 3: A maximum network error of
0.01, the momentum rate and learning rate
range between 0.1 and 0.9 are used in the
training.
Step4: The trained network obtained in
step 3 has been tested with randomly
selected data from the 40% testing sets.
VI. CONCLUSION AND DIRECTION
A. Conclusion
The following Table IV displays the
Actual value Vs. Predicted value against
the companies under BSE on randomly
selected dates from the 40% testing set.
445
Volume: 02, December 2013, Pages: 440-449
International Journal of Computing Algorithm
Table V displays the performance of the
network adopted in the study.
TABLE IV. ACTUAL VALUE Vs. PREDICTED VALUE OF STOCK PRICES
(NN1-3-16-1)
(NN2-3-6-1)
Actual
value Predicted Forecasting Predicted Forecasting
(Rs.)
value
Error (%)
value
Error (%)
29/4/2013 1370.70 1452.00
5.93
1423.96
3.89
Tata
Consultancy
5/7/2013 1527.35 1604.98
5.08
1551.32
1.57
Services Ltd
6/8/2013 1870.50 1787.97
4.41
1742.31
6.85
0.39
2784.85
0.30
18/2/2013 2776.55 2787.30
Infosys
8/4/2013 2833.15 2843.44
0.36
2751.08
2.90
Technologies
Ltd.
17/9/2013 3023.35 2909.26
3.77
3348.80
10.76
6/3/2013 1807.70 1766.09
2.30
1801.16
0.36
Dr. Reddy’s
6/5/2013 1992.75 1934.31
2.93
1950.08
2.14
Laboratories
Ltd
6/8/2013 2197.65 2010.44
8.52
2104.98
4.22
Sun
13/03/2013 418.88
420.36
0.35
422.63
0.90
Pharmaceutical 3/4/2013
432.30
403.12
6.75
427.41
1.13
Ltd
14/6/2013 476.42
478.94
0.53
521.43
9.45
Company
Name
Date
TABLE V. EVALUATION OF THE NEURAL NETWORK
Input
Parameters
(Previous
three
closing
prices)
Neural
Network
Architecture
p t 1
(NN1-3-161)
pt 2
pt 3
(NN2-3-6-1)
Forecasting
Performance
Tata
Consultancy
Services Ltd
Infosys
Technologies
Ltd
Dr. Reddy’s
Laboratories
Ltd
Sun
Pharmaceutical
Ltd
MAPE
MAD
RMSE
MAPE
MAD
RMSE
5.14
80.49
80.51
4.10
68.47
81.33
1.51
45.04
66.43
4.65
138.61
193.84
4.58
95.62
115.68
2.24
47.29
59.02
2.54
11.06
16.93
3.83
17.88
26.23
The graphical representations of actual and
predicted prices of the companies are
shown in Fig. 1, Fig. 2, Fig. 3 and Fig. 4.
The horizontal axis in the graphs denotes
randomly selected dates taken in growing
order for forecasting and the vertical axis
denotes the respective stock price.
Figure 1. Actual Vs. Predicted stock prices
of Tata Consultancy Services Ltd
Integrated Intelligent Research (IIR)
446
Volume: 02, December 2013, Pages: 440-449
International Journal of Computing Algorithm
Figure 3. Actual Vs. Predicted stock prices
of Dr. Reddy’s Laboratories Ltd
Figure 2. Actual Vs. Predicted stock prices
of Infosys Technologies Ltd
Figure 4. Actual Vs. Predicted stock prices
of Sun Pharmaceutical Ltd
Comparing the performance of the two
networks, NN1 performs better than NN2
by observing the MAPE results and are
shown in Fig. 5, Fig. 6 and Fig. 7.
Figure 5. MAPE comparison for NN1 and NN2
Integrated Intelligent Research (IIR)
447
Volume: 02, December 2013, Pages: 440-449
International Journal of Computing Algorithm
Figure 6. MAD comparison for NN1 and NN2
Figure 7. RMSE comparison for NN1 and NN2
A highly flexible non-linear modeling
technique ANN has been implemented to
forecast the stock prices of selected sectors
under Bombay Stock Exchange. Two
different neural networks used in this
study namely NN1 (3-16-1) and NN2 (3-61) where sigmoid function is considered as
transfer function. The architectures are
tested with the testing data set and the
results are predicted. The input data used
in the model is the preprocessed historic
closing prices of the companies. The
model predicts the closing price if three
consecutive trading days’ stock prices are
supplied to the model. In a highly volatile
market like Indian Stock Market, if the
prediction of the direction of the market is
possible with fairly high accuracy it will
guide the investors to reap benefits. The
predicted results show that artificial neural
network has been able to predict stock
prices and ensure that it is suitable for
forecasting with better accuracies.
Limitations and Future Scope of
Improvement In this study only the
historic prices of stock has been used for
Integrated Intelligent Research (IIR)
the prediction problem with artificial
neural network. In order to improve the
accuracy of the model, macroeconomic
factors and international stock market data
can also be used as input variables.
Technical analysis indicators can also be
used in the input variables and can be
checked for improvement in the
performance of the network.
References
[1]A. Abhyankar, L.S. Copeland and W.
Wong, “Uncovering nonlinear structure in
real-time stock market indexes: The S&P
500, the DAX, the Nikkei 225 and the
FTSE-100”, Journal of Business &
Economic Statistics, No 15, pp 1-14, 1997.
[2]M.T. Hagan, H.B. Demuth and., M.
Beale, “Neural network design”. PWS
Publishing Company, Boston, 1996.
[3]M.R. Hassan and B. Nath,, “Stock
Market Forecasting using Hidden Markov
Model: A new approach”, Proceedings of
5th International conference on Intelligent
systems Design and Applications, 2005.
448
Volume: 02, December 2013, Pages: 440-449
International Journal of Computing Algorithm
[4]H. White, “Learning in neural
networks: A statistical perspective”,
Neural Computation. No. 4, pp. 425-464,
1989.
[5]R. L. Wilson and R. Sharda,
:Bankruptcy Prediction Using Neural
Networks”, Decision Support Systems,
No. 11, pp. 545-557, 1994.
[6]G. Zang, B.E. Patuwo and M.Y. Hu,
“Forecasting
with
artificial
neural
networks: The state of the art”,
International Journal of Forecasting, 14,
pp. 35-62, 1998.
[7]A. Lapede and R. Farber, “Nonlinear
signal processing using neural networks:
prediction
and
system
modeling”,
Technical Report LA-UR-87-2662, Los
Alamos National Laboratory, Los Alamos,
NM, 1987.
[8]M. Kaastgra and M. Boyd, “Designing
a neural network for forecasting financial
and
economic
time
series”,
Neurocomputing, No. 10, pp. 215-236,
1996.
[9]M. Pring, “Martin Pring’s Introduction
to Technical Analysis”, Singapore:
McGraw-Hill, 1999.
[10]U.
BenZion,
P.
Klein,
Y.
Shachmurove and J. Yagil, “Efficiency
Differences Between the S&P 500 and the
Tel-Aviv 25 indices: A Moving Average
Comparison”. International Journal of
Business, Vol. 8, No.3, pp. 267-284, 2003.
[11]A. Dzikevicius, S. Saranda, and A.
Kravcionok, “The Accuracy of Simple
Trading Rules in Stock Markets”,
Economic and Management, No.15, pp.
910-916, 2010.
[12]B. Vanstone and G. Finnie,
“Combining Technical Analysis and
Neural Networks in the Australian
Stockmarket”.
Bond
University
ePublications@bond.
Information
technology papers, 2006.
[13]W. Huang, Y. Nakamori and Y. Wang,
“Forecasting stock market movement
direction with support vector machine.
Computers and Operations Research”, No.
32, pp. 2513-2252, 2005.
[14]T. Mitchell, “Artificial Neural
Networks: Based on Machine Learning”,
Mc Graw Hill, 1994.
[15]Simon Haykin, “Neural Network A
comprehensive
Foundation”,
Second
edition, Prentice Hall, pp. 161-173, 1998.
[16]S. Chaigusin, C. Chirathamjaree and
J.M. Clayden, “The Use of Neural
Networks in the Prediction of Stock
Exchange of Thailand (SET) Index”,
Proceedings of International Conference
on Computational Intelligence
for
Modelling, Control and Automation,
Vienna, Austria, pp. 670-673., 2008.
[17]S. Saigal and D.
Mehrotra,
“Performance Comparison of Time Series
Data Using Predictive Data Mining
Techniques”, Advances in Information
Mining, No.1, pp. 57-66, 2012.
[18]http://neuroph.sourceforge.net
[19]http://in.finance.yahoo.com
[21]M.J.C. Hu,. “Application of the
adaline system to weather forecasting”.
Master thesis, Technical Report 6775-1.
Stanford Electronic Laboratories, Stanford,
CA, June, 1964.
[22]W. Huang, Y. Nakamori and Y. Wang,
“Forecasting stock market movement
direction with support vector machine.
Computers and Operations Research”, No.
32, pp. 2513-2252, 2005.
[23]Z.H. Khan, T.S. Alin and M.A.
Hussain, “Price Prediction of Share
Market using Artificial Neural Network
(ANN)”,
International
Journal
of
Computer Applications, 22, No 2, pp. 4247, 2011.
[24]M. Majumder and M.A. Hussain, “A.
Forecasting of Indian Stock Market Index
using Artificial Neural Network” ,
unpublished article.
[25]S.
Rajasekaran
and
G.A.
Vijayalakshmi, “Neural Networks, Fuzzy
logic and Genetic Algorithms”, PHI, 2003.
[26]R. Lawrence, “Using Neural Networks
to Forecast Stock Market Price", 1997.
Integrated Intelligent Research (IIR)
449