See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/252016485

An ANFIS model for stock price prediction: The case of Tehran stock
exchange

Conference Paper · June 2011

DOI: 10.1109/INISTA.2011.5946124

CITATIONS READS

16 224

2 authors:

Akbar Esfahanipour Parvin Mardani

Amirkabir University of Technology Amirkabir University of Technology
45 PUBLICATIONS   479 CITATIONS    1 PUBLICATION   16 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Volatility Modeling and Forecasting View project

Behavioral Portfolio Selection View project

All content following this page was uploaded by Akbar Esfahanipour on 16 June 2015.

The user has requested enhancement of the downloaded file.

 An ANFIS Model for Stock Price Prediction: The
Case of Tehran Stock Exchange
Akbar Esfahanipour, Parvin Mardani
Department of Industrial Engineering and Management Systems, Amirkabir University of Technology, Tehran, Iran
esfahaa@aut.ac.ir, mardaniparvin@aut.ac.ir
Abstract— The main purpose of forecasting in financial markets
is to estimate future trends and to reduce risks of decision
making. This research suggests an ANFIS model to improve
prediction accuracy in stock price forecasting. For doing so, we
applied fuzzy subtractive clustering for structure identification of
our ANFIS model. We implemented the proposed model for
predicting Tehran Stock Exchange Price Index (TEPIX) using a
dataset including TEPIX data from 25 March 2001 until 25
September 2010. To demonstrate the advantages of this model,
first we compared our results with an Artificial Neural Network
(ANN) model of type Multi Layer Perceptron (MLP). Then, we
compared our results with ANFIS models using grid partitioning
and Fuzzy C-Mean (FCM) clustering. The comparative results
show the superiority of our proposed ANFIS model against ANN
model and ANFIS models with no clustering and FCM
clustering.
Keywords- Stock price prediction; ANFIS; fuzzy subtractive
clustering; Artificial Neural Network; Tehran Stock Exchange
I. INTRODUCTION
Followers of random walk theory and the Efficient Market
Hypothesis (EMH) believe that stock price in an efficient
market is unpredictable. However, adherents of technical
analysis, fundamental analysis, nonlinear dynamics and chaos
theory believe on the predictability of stock price and they have
provided several methods for this purpose.
Recent advances in soft computing techniques provide
suitable tools to predict chaotic environments such as stock
market with taking into account their nonlinear behaviors [1].
Fuzzy Inference Systems (FISs), Artificial Neural Networks
(ANNs) and heuristic techniques such as Genetic Algorithm
(GA) are principle constituents of soft computing. The hybrid
of these techniques is quite effective which synergistically
exploits the advantages of two or more of these areas [2].
Neuro- fuzzy systems are an example of these combinatorial
methodologies. Adaptive Network based Fuzzy Inference
System (ANFIS), as presented by Roger Jang in 1993, is one of
the most powerful structures that widely used in financial
markets.
Although neuro- fuzzy systems have the advantages of
faster convergence, higher accuracy of prediction, and needless
to make any assumptions about statistical properties of data;
defining their structures is not a simple task. Structure
identification relates to effective partitioning of the input space
978-1-61284-922-5/11/$26.00 ©2011 IEEE
44
to generate an FIS with the minimum number of rules to
distinguish data behavior to achieve promising results [3].
In this perspective, the purpose of this research is to present
a successful application of ANFIS using fuzzy subtractive
clustering to determine appropriate number of rules and
membership functions (MFs) in the first order Takagi-Sugeno-
Kang (TSK) fuzzy rule base for stock price prediction.
Additionally, performance of our proposed ANFIS model has
been compared with ANFIS models based on different
clustering methods, grid partitioning as well as an Artificial
Neural Network (ANN) model of type Multi Layer Perceptron
(MLP).
The rest of this paper has been organized as follows.
Section II provides the relevant studies. Section III reviews the
overall structure and learning procedure of ANFIS besides
fuzzy subtractive clustering method. The utilized performance
measures for evaluating our model are described in section IV.
Section V presents the structure of our proposed model.
Section VI describes implementation of the proposed approach
in Tehran Stock Exchange (TSE) and comparison results of the
models. Finally, some concluding remarks are pointed out in
section VII.
II. LITERATURE REVIEW
Two studies of [4-5] provided several evidences for
predictability of stock price in Tehran Stock Exchange (TSE)
as well as its chaotic and nonlinear behavior in this emerging
market. Therefore, it is reasonable to apply an ANFIS model,
as a more suitable approach, for prediction of stock price in this
market rather than the traditional forecasting methods which
were not efficient enough.
Table 1 summarizes characteristics of recent studies that
utilized an ANFIS model for stock price prediction in
comparison with this study. Some of these studies compare
their derived ANFIS model with other techniques including
Support Vector Machine (SVM) and various types of Artificial
Neural Networks such as Back Propagation Neural Network
(BPNN), Multi Layer Perceptron (MLP), Different Boosting
Neural Network (DBNN) and Radial Basis Function (RBF).
All these researches justify the high potential of ANFIS based
modeling in stock market prediction.
Reference [9] proposed a volatility model based on fusion
ANFIS models with Fuzzy C-mean (FCM) and subtractive
clustering for partitioning the universe of discourse.
 TABLE I. RECENT STUDIES THAT UTILIZED AN ANFIS MODEL FOR STOCK PRICE PREDICTION

Fuzzy clustering Performance

Reference Input Data Sample Size Model Comparisons Case Study
method Measures
7, 4 years of RMSE, MAPE, NASDAQ and S&P
open, close, high and
[6] NASDAQ and ANN, SVM, DBNN maximum absolute CNX NIFTY stock
low Index price
NIFTY percentage error indices
fitting error,
Index values of five 244 forecasting error, Shanghai Stock
[7]
previous days observations ratio of correct Market Index
tendency
B&H and thirteen Some stocks of
4775 reported soft computing Hit, RMSE, MAE, Athens and New
[8] daily stock price
observations models such as RBF, MSE York Stock
MLP, FIS and etc. Exchange (NYSE)
NASDAQ, DJ and Two reported fuzzy Taiwan Stock
c-mean,
[9] TAIEX values of one 7 years models in the literature RMSE Exchange Index
subtractive
previous day for time series prediction (TAIEX)
614, 479 Multi regression analysis,
Six days moving Taiwan and Tehran
observations of BPNN and one reported
[10] c-mean average and six days MAPE Stock Exchange
Taiwan and TSK model in the
bias Indices
Tehran Index literature
Six monthly
macroeconomic RMSE, R2,
228 Istanbul Stock
[11] indicators and three coefficient of
observations Exchange return
indices of DJI, DAX, variance
and BOVESPA
BPNN, SVM and two
close, high and low 1000
[12] hybrid models based on MSE S&P500 Index
Index price observations
ANFIS, BPNN and SVM
MLP, ANFIS based on Tehran Stock
c-mean, TEPIX values of four 2297 RMSE, MAE,
This study two clustering methods Exchange Price
subtractive previous days observations MAPE, NMSE, R2
and grid partitioning Index (TEPIX)

They put into practice both these clustering methods in
implementation steps of their model and selected the one with
better results. They concluded that the best empirical results
had been obtained by ANFIS with subtractive clustering.
Reference [10] presented adaptive neuro-fuzzy model based on
TSK fuzzy inference system and applied FCM clustering for
identifying number of rules that showed higher performance in
comparison with multiple regression and BPNN.
Collazo et al. accomplished a comparison between FCM
and subtractive clustering as a basis of fuzzy model
identification algorithm [13]. They used both generated FISs in
prediction air pollution at northwest of England. Their study
asserted the feasibility of both methods to forecast behavior of
complex dynamic systems.
clustering is somewhat preferred and have had conductive
applications as modeling the solar power plant [14], PI type
fuzzy logic controller [18], Twin Rotor MIMO Systems
(TRMS) control [17] and prediction chaotic time series of
traffic flow [16].
III. THEORETICAL BASICS
A. ANFIS Architecture and Learning Procedure
To give a concise explanation of ANFIS as a basis of our
proposed model, we assume first order TSK fuzzy inference
2
system with input x = ( x1 , x 2 ) ∈ U ⊂ R and two fuzzy
rules as:

Pereira and Dourado constructed a neuro-fuzzy system
based on RBF and SVM that take advantage of subtractive
clustering for complexity reduction [14]. They contend that
using the subtractive clustering is more efficient than the FCM
in the sense that the number of clusters does not need a prior
specification while the quality of the FCM strongly depends on
the choice of the number of centers and initial cluster positions.
FCM and fuzzy subtractive clustering require setting of
prerequisite parameters as the number of clusters and clustering
radius, respectively. Pre-assigned cluster radius implicitly
determines the number of clusters in later technique as
mentioned in [15-17]. On the other hand, we don’t have any
idea or prior knowledge about the number of clusters in a given
data set in most cases. In such conditions, fuzzy subtractive
If x1 is C1l and x 2 is C 2l then y l = c0l + c1l x1 + c 2l x 2 , l=1, 2
l l
Where Ci represents fuzzy sets or linguistic values, ci shows
constant coefficients, and l is the number of rules. The
following is a layer by layer description of ANFIS architecture
based on the above-mentioned TSK system.
• Layer1: There are two inputs and two Membership
Functions (MFs) associated to each one and so this
layer contains four adaptive nodes. The output of a
node μ C l ( x ) is the MF of Ci . We can choose the MF
l
i
as any continuous and piecewise differentiable
functions between 0 and 1 such as Gaussian function

45
 of (1) which is used in this study. It has two parameters
of {ai, bi} that are related to antecedent parameters.
⎡ x − bi 2 ⎤
μ Cil ( x) = exp ⎢− ( ) ⎥ (1)
⎣ ai ⎦ acceptance ratio of ε and rejection ratio of ε ′ . Cluster
• Layer2: It has two fixed nodes that every node
generates the weight of the l-th rule by (2).
wl = ∏ i =1,2 μC l ( xi ) = μC l ( x1 ) × μC l ( x2 ) (2)
i 1 2
• Layer3: the number and type of nodes in this layer is
the same as the previous layer. The output of a fixed
node is a normalized weight which is defined as (3).
l
w = wl ( w1 + w2 ) (3)
• Layer4: There are two adaptive nodes that calculate
rule outputs based on the consequent parameters and
then multiply them by the related normalized weight,
l l
as a result we have w y that is sent to the next layer
from each node. Each rules output is obtained by (4).
y l = c0l + c1l x1 + c2l x2 (4)
• Layer5: The final output of the network can be
achieved in this layer. The single node in this last layer
calculates f ( x ) ∈ V ⊂ R that is equivalent to the
first order TSK fuzzy inference system output. It is
obtained by the summation of incoming signals from
layer 4 as shown by (5).
f ( x) = ∑ w y l = w y + w y 2
l 1 1 2
(5)
In this study, parameters recalibration has been performed
through combination of Error Back Propagation (EBP) and
Least Square (LS) estimates that is known as hybrid learning
rule. In MATLAB implementation, we can control step size,
step size decrease rate, and step size increase rate as well as
error tolerance and the number of training epochs. For more
details of ANFIS architecture and hybrid learning please refer
to [3].
B. The Fuzzy Subtractive Clustering
As stated in [19], the Fuzzy subtractive clustering was
introduced by Chiu which is an expansion of the grid-based
mountain clustering method that proposed by Yager and Filev.
This Algorithm was designed for estimating the number of
cluster centers in a set of data, particularly for high dimensional
problems [19]. Assuming there are n data points in an m
46
dimensional space and let k as counter of cluster numbers, this
algorithm does the followings:
• Step1: Rescaling data into [0, 1] in each dimensions.
• Step2: Setting cluster radius ra, quash factor,
radius specifies a cluster center’s range of influence in
each of the data dimensions. Quash factor is used in
consequent steps to determine the neighborhood of a
cluster center within the existence of other cluster
centers. We can also obtain upper and lower thresholds
for potential cluster center by means of acceptance and
rejection ratios in the following steps.
• Step3: Each data point is assumed as a potential
cluster center. If {x1, x2, …, xn} shows normalized
points, potential value of i-th data as a cluster center is
defined by (6).
⎡− x − x 2
⎤
Pi = ∑ j =1 exp ⎢ ⎥
n i j
(6)
⎢ (ra 2) 2
⎥
⎣ ⎦
• Step4: The data point xt with the maximum potential Pt
is considered as the first cluster center, k=1, hence we
show it as ( xt(1) , Pt (1) ) .
• Step5: The first cluster center and all data points in its
neighborhood must be removed and the potential
values of the remaining points must be revised. In
other words the new potential value of xi excludes the
influence of the first cluster center. In general, equation
(7) shows the revised potential value after obtaining k-
th cluster center where rb = quash factor * ra.
⎡ − x − x (k ) 2
⎤
exp ⎢ ⎥
new old (k ) i t
Pi = Pi − Pt (7)
⎢ (rb 2) 2 ⎥
⎣ ⎦
• Step6: The data point xt with the highest revised
potential value of Pt is a candidate for the next cluster
center. If Pt εPt (1) or ⎡(dmin ra ) + ( Pt Pt (1) ) ⎤ ≥ 1 where
⎣ ⎦
dmin is the minimum distance between xt and all the
previous cluster centers, we accept xt as a new cluster
center and set cluster counter as k=k+1 so we have
( xt( k +1) , Pt ( k +1) ) , then we go to the previous step. In
situation that any of these conditions are not satisfied,
the next data with the highest potential value will be
tested. This process will be continued until Pt ≺ ε ′Pt (1) .
This condition is used as stopping criteria for acquiring
new cluster center in this algorithm.
Next, each cluster center can be considered as a basis for
constructing a rule so that initial parameters of a rule can be
identified [13, 15, 19]. Applying this process in our proposed
model, we obtain appropriate number of rules and membership
functions for each input by finding the number of clusters in
input-output training set besides initial rule parameters. The
number of clusters is equal to the number of fuzzy rules.
IV. PERFORMANCE MEASURES
We applied the commonly used performance measures in
forecasting problems to compare and evaluate the accuracy of
our model. These performance measures are Root Mean Square
Error (RMSE), Mean Absolute Error (MAE), Mean Absolute
Percentage Error (MAPE), Normalized Mean Square Error
(NMSE) and coefficient of determination (R2) which are
calculated by (8) – (12) where Q is the number of testing
patterns, dq is the actual output for pattern q, zq is the model
output for pattern q, and d is the average of all actual outputs
related to Q testing patterns.
∑
Q
RMSE = q =1
( d q − zq ) 2 Q
MAE = ∑ q =1 d q − zq Q
Q
MAPE = (100 Q) × ∑ q =1 (d q − zq ) d q
Q
NMSE = ∑ q =1 (d q − zq ) 2 ∑
Q Q
q =1
(d q − d ) 2
R 2 = 1 − ⎡ ∑ q =1 (d q − zq ) 2 ∑ (d q − d ) 2 ⎤ (12)
Q Q
⎣ q =1
V. THE PROPOSED MODEL FOR STOCK PRICE PREDICTION
The following steps show the implementation process of
the proposed model.
• Step1: Collecting data.
• Step2: Determining training, validation and testing
input- output sets.
• Step3: Generating initial first order TSK fuzzy
inference system through fuzzy subtractive clustering.
Step3-1: Estimating clusters.
Step3-2: Extracting rule and initial rule parameters.
• Step4: Tuning antecedent and consequent parameters
by ANFIS hybrid learning algorithm.
After setting training parameters, learning algorithm
will be continued to reach the specified error goal or
maximum number of epochs.
• Step5: Calculating performance measures.
• Step6: Setting cluster radius and step size and then
running the model.
• Step7: Selecting the model with the minimum
validation error.
By using clusters information and neural methodology in
ANFIS structure, this procedure establishes the first order TSK
fuzzy inference system that models the data behavior in a better
way. The flowchart of our proposed model is shown in Fig. 1.
VI. IMPLEMENTATION OF THE PROPOSED MODEL IN TSE
In this section, we describe the best structure obtained for
the proposed model to forecast Tehran Stock Exchange Price
(8) Index (TEPIX). Comparative studies are represented for a more
accurate assessment.
A. Input Data
(9) Four lags of TEPIX, that is xt-1, xt-2, xt-3, and xt-4, are used as
input variables and the model predicts one step ahead of Index
x̂t by finding the function f in xˆt = f ( xt −1 , xt − 2 , xt −3 , xt − 4 ) .
In order to test generalization capability of the FIS at each
(10) epoch, we separated our data into training, validation and test
sets as 70%, 15% and 15%, respectively. Our training and
validation sets include 1957 data sets that encompass TEPIX
values from 25 March 2001 to 4 May 2009. Test set contains
(11) 340 data sets from 5 May 2009 to 25 September 2010.
B. Numerical Results and Model Comparisons
Most fitting structure of our proposed model to predict
TEPIX has a first order TSK fuzzy inference system with four
⎦ inputs, one output and four fuzzy rules. Each input variable has
four Gaussian membership functions and each rules output is a
linear combination of input variables plus a constant term.
The result of fuzzy subtractive clustering was to separate
input-output training set into four clusters where cluster radius
of 0.45 had been specified for all data dimensions, since it gave
the best generalization power. Quash factor, acceptance ratio
and rejection ratio were set as 1.25, 0.5 and 0.15, respectively.
These values are default values of MATLAB 7.11.0. ANFIS
hybrid learning tuned antecedent and consequent parameters
within 2000 epochs with error tolerance of zero, step size of
0.1, step size increase rate of 1.1 and step size decrease rate of
0.9.
To set cluster radius and according to [16], we consider the
restriction of 0.1 ≤ ra ≤ 0.5. In fuzzy subtractive clustering,
range of influence strongly affects the number of clusters (i.e.,
number of rules) and performance level of the model.
47
 activation function in output layer. Faster training techniques
Collecting Data including heuristic techniques and numerical optimization
techniques were employed to improve Error Back Propagation
(EBP) algorithm. The minimum error achieved with Levenberg
Determine training, validation and testing
input-output sets Marquardt algorithm in 5000 epochs. To acquire appropriate
results in the neural network model, we performed pre-
processing on the inputs and outputs to normalize the mean and
standard deviation. In this situation, the output of the network
STRUCTURE IDENTIFICATION is trained to produce zero mean and unity standard deviation.
Set cluster radius, quash factor, acceptance Therefore, we used post-processing to convert these outputs
and rejection ratios back to the original scale.

65.5
Perform fuzzy subtractive clustering on
input- output training set 65
5 rules
10 rules
5 rules
64.5
Generate initial FIS 5 rules
64 11 rules

RMSE
63.5
6 rules 3 rules
ANFIS STRUCTURE 63

Set number of epochs, step size, step size 62.5 4 rules

decrease and increase rate
62

Adjust antecedent and consequent parameters 61.5

of rules by hybrid learning 61
4 rules
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Cluster radius for each data dimensions

Calculate performance measures

Figure 2. The comparison of our model results regarding to the change of
cluster radius
Reset cluster radius and step size
and run the model
4
x 10
2
Choose the model with
minimum RMSE validation
1.5
TEPIX

Finish

1
Figure 1. Flowchart of the proposed model
Actual Model
Output Output
For showing this fact, we have performed a comparison 0.5
between number of rules, RMSE validation against the change 0 50 100 150 200 250 300 350
of cluster radius. As shown in Fig. 2, larger range of influence
leads to fewer clusters, then fewer number of rules and vice
versa. 500
Since we considered validation set, the final FIS structure is
the one associated with the minimum validation error. This
structure is used for calculating network output per training,
Error

validation and testing sets. Fig. 3 shows the comparison of 0

actual output (i.e., solid line) and the proposed model’s output
(i.e., dashed line) for testing data.
For more investigation on response of our proposed model -500
we employed ANN techniques to forecast TEPIX. According 0 50 100 150 200 250 300 350
to Universal Function Approximation theorem, we established Data Number
an MLP and the best results achieved by the one that has four
nodes in input layer, four Sigmoidal neurons with Tansig
activation function in hidden layer and one node with linear Figure 3. The comparison of actual and model output for testing set

48
 TABLE II.
No. of No. of
Model RMSE
MFs Rules
Multi Layer Perceptron - - 80.0096
ANFIS with grid partitioning 2 16 178.8469
ANFIS with FCM clustering 4 4 80.0793
Our proposed ANFIS model 4 4 78.0690
To evaluate the effectiveness of subtractive clustering on
performance of ANFIS model, we implemented ANFIS
without clustering and with FCM clustering. Table II shows
numerical results of the best observed performance for MLP,
ANFIS with grid partitioning, ANFIS with FCM clustering and
our proposed model.
Our empirical results confirm the positive impact of
clustering on increasing strength of ANFIS. These techniques
reduced the number of rules with higher level of generalization
performance. The superiority of our proposed model in
comparison with neural network model justifies application of
ANFIS for stock price prediction rather than MLP model.
VII. CONCLUSION
Combining a fuzzy inference system into an ANFIS
structure, we take advantage of neural methodology to train a
fuzzy system that intensifies its capability for dealing with
nonlinearity and imprecision in addition to its easier
implementation. In this research, we synthesized an ANFIS
model with fuzzy subtractive clustering aimed at some
dimensionality reduction to construct a first order TSK fuzzy
inference system with the minimum number of rules.
We implemented our proposed model in Tehran Stock
Exchange (TSE) to investigate its accuracy. More than 8 year’s
TEPIX values were utilized and the results were compared
with the best obtained results of ANFIS with no clustering,
ANFIS with FCM clustering and Multi Layer Perceptron with
Levenberg Marquardt training algorithm. Based on our
comparative results, clustering has a positive influence on the
performance of ANFIS model. This advantage is more
observable when the large number of input data leads to
explosion of rule numbers in constructing an ANFIS structure
based on grid partitioning. The utilized performance measures
confirmed higher forecasting accuracy of our model in
comparison with the best results of MLP and ANFIS with
FCM clustering. Our results also show the predictability of
stock price trend using historical prices that violates weak form
of Efficient Market Hypothesis (EMH) in TSE.
REFERENCES
[1] G. S. Atsalakis, and P. K. Valavanis, “Surveying stock market
forecasting techniques- part II: soft computing methods,” Journal of
Expert System with Applications, vol.36, 2009, pp.5932-5941.
[2] F. Karray, and C. D. Silva, Soft Computing and Intelligent System
Design, Addison Weseley, 2004.
[3] J. R. Jang, “ANFIS: Adaptive network based fuzzy inference system,”
IEEE Trans. On Systems, Man and Cybernetics vol. 23, no. 3, 1993, pp.
665-685.
49
View publication stats
NUMERICAL RESULTS
MAE MAPE NMSE R2
54.6414 0.4133 0.0009957 0.9990
91.7062 0.6313 0.0050 0.9950
53.9239 0.4103 0.0009975 0.9990
53.3059 0.4050 0.0009480 0.9991
[4] H. Khalouzadeh, and A. Khaki, “Evaluating stock price prediction
methods and representing a nonlinear model based on neural networks,”
(In persian), Economic Researches (published by Tehran University
Faculty of Economics), no. 64 , 2003, pp. 43-85.
[5] Kh. Nasrollahi, S. Samadi, and R. Saghafi, “An appraisal of
predictability in TEPIX,” (In persian), Quarterly Journal of Tehran
Securities& Exchange Organization, vol. 2, no. 6 , 2009, pp. 5-30.
[6] A. Abraham, N. S. Philip, and P. Saratchandran, “Modeling chaotic
behavior of stock indices using intelligent paradigms,” 2004, available at
http://arxiv.org/abs/cs.
[7] R. J. Li, and Zh. B. Xiong, “Forecasting stock market with fuzzy neural
networks,” IEEE, [4th International Conf. on Machine Learning and
Cybernetics, pp. 3475-3479, 2005].
[8] G. S. Atsalakis, and K. P. Valavanis, “Forecasting stock market short
term trends using a neuro-fuzzy based methodology,” Journal of Expert
Systems with Applications, vol. 36, no. 7, 2009, pp.10696-10707.
[9] Ch. H. Cheng, L. Y. Wei, and Y. Sh. Chen, “Fusion ANFIS models
based on multi- stock volatility causality for TAIEX forecasting,”
Journal of Neurocomputing, vol. 72, no. 16-18, 2009, pp. 3462-3468.
[10] A. Esfahanipour, and W. Aghamiri, “Adapted neuro-fuzzy inference
system on indirict approach TSK fuzzy rule base for stock market
analysis,” Journal of Expert Systems with Applications, vol. 37, 2010,
pp. 4742-4748.
[11] M. A. Boyacioglu, and D. Avci. “An adaptive Network Based Fuzzy
Inference System (ANFIS) for the prediction of Stock Market Return:
The Case of Istanbul Stock Exchange.” Journal of Expert Systems with
Applications 37, no. 12, 2010, pp. 7908-7912.
[12] F. Zhai, Q. Wen, Z. Yang, and Y. Song. “Hybrid forecasting model
research on stock data mining,”IEEE, [4th International Conf. on New
Trends in Inf. Science and Service Science (NISS), pp. 630-633, 2010].
[13] J. I. Collazo Cuevas, M. A. Aceves Fernandez, E. Gorrostieta Hurtado,
J. C. Pedraza Ortega, A. Sotomayor Olmedo, and M. Delgado Rosas.
“Comparison between fuzzy c-mean clustering and fuzzy clustering
subtractive in urban air pollution,” IEEE, [20th international conference
on Electronics, Communications and Computer (CONIELECOMP), pp.
174-179, 2010].
[14] C. Pereira, and A. Dourado, “Neuro-fuzzy systems complexity reduction
by subtractive clustering and support vector learning for nonlinear
process modeling,” eunite, 2001.
[15] A. Ghodsi, “Efficient parameter selection for system identification,”
IEEE, [Annual Meeting of the Fuzzy Information (NAFIPS'04), pp. 750-
755, 2004].
[16] M. b. Pang, and X. P. Zhao, “Traffic flow prediction of chaos time series
by using subtractive clustering for fuzzy neural network modeling,”
IEEE, [Second International Symposium on Intelligent Information
Technology Application, pp. 23-27, 2008].
[17] T. Sh. Mahmoud, M. H. Marhaban, and T. S. Hong, “ANFIS controller
with fuzzy subtractive clustering method to reduce coupling effects in
twin rotor MIMO systems (TRMS) with less memory at time usage,”
IEEE, [International Conf. on Advanced Computer Control, pp. 19-23,
2008].
[18] S. Chopra, R. Mitra, and V. Kumar, “Identification of rules using
subtractive clustering with application to fuzzy controllers,” IEEE, [The
Third International Conf. on Machine Learning and Cybernetics
Shanghai, 4125-4130, 2004].
[19] S. Chiu, “Method and software for extracting fuzzy classification rules
by subtractive clustering,” IEEE, 1996, pp. 461-465.