SCIENCE ROAD

Journal

SCIENCE ROAD JOURNAL

Year: 2013 Volume: 01 Issue: 03 Pages: 31-41

In research of optimal model for predicting bankruptcy

Sajad Abdipour
M.A of Accounting, USB University, Iran
Mojtaba Akbarpour
M.A of Accounting, USB University, Iran
Amir Shams Koloukhi
Young Researchers and Elite Club, Torbat-e-Jam Branch, Islamic Azad University, Torbat-e-
Jam Branch, Iran
Hossein Parsian
Young Researchers and Elite Club, Torbat-e-Jam Branch, Islamic Azad University, Torbat-e-
Jam Branch, Iran

Abstract:
In this study we tried to compare two models in order to identify optimal neural networks models in predicting bankruptcy.
Multi-layered perceptron(MLP) because of easy training and high efficiency and also integrated multi-layered perceptron (most
used neural network in predicting bankruptcy) with new imperialist competitive algorithm(MLP-ICA) are used in this study.
Research sample consist of 70 bankrupts and non-bankrupt company in 2001-2009 and in listed firms of Tehran Stock
Exchange. Results indicate that notwithstanding their good performance, they aren’t any significant difference in their
performance.

Keywords: Bankruptcy, Neural Network, imperialist competitive algorithm

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
31
 SCIENCE ROAD JOURNAL
1. Introduction

Predicting companies’ financial crisis is a subject which has attracted attentions of many accounting researchers
towards it. The passage of studies related to bankruptcy prediction has experienced a rapid growth during recent
decades. This passage has moved from uinvariate analyses in 60s (Beaver, 1968) towards applying various
optimization algorithms such as artificial neural networks, genetic algorithm, particle swarm algorithm, etc.

Artificial neural network is an attempt to assimilate function of man’s brain and first were invented by McCulloch–
Pitts (1947) and after that was used in various fields. Recent studies on artificial neural networks (ANN) indicates
that due to having nonlinear, nonparametric characteristics, and learning ability, artificial neural networks are
considered as powerful tools for patterns recognition and grouping.

Neural networks have been applied in different commercial areas including bankruptcy prediction (Tseng and Hu,
2010), credit rating (Jensen, 1992; Piramuthu et al., 1994)bond rating(Kim et al., 1993; Dutta et al., 1994),
predicting future prices (Grudnitski and Osburn, 1993), analyzing financial statements (Kryzanowski and Galler,
1995). However, in spite of various utilizations of this model in different commercial areas, the literature of
bankruptcy prediction has devoted more parts of these researches (Wong and Selvi, 1998). The common point
among most of researches conducted upon bankruptcy prediction using neural networks is the subject of training
neural networks, so that most of researchers mentioned multi-layered perceptron (MLP), for which Back
propagation Algorithm has been utilized for training (Anandarajan et al., 2004; Lee, 2001; Lee et al., 2005; O’Leary,
1998; Tam and Kiang, 1992; Tseng and Hu, 2010; Yang et al., 1999; Zhang et al., 1999).

One of features that discriminates neural network from other models utilized for predicting bankruptcy is its learning
ability. Learning ability is the ability for regulating network’s parameters on the time direction with different
conditions, with an aim that trained (taught) network for a special situation can maintain its efficiency in spite of
small environmental changes. However, despite continuous utilizations of Back propagation Algorithm, the presence
of some weaknesses such as entrapment in local minimum, low convergence speed, excessive dependency on
network effectiveness for selecting initial weights (Hagan et al., 1996) are among criticism considered upon this
algorithm. However simplicity of this model caused its extensive use by researcher.

Therefore, it is intended in present study besides emphasis on characteristics and advantages of this algorithm, its
performance in developing countries is examined. The selected area for this examining this model is Tehran stock
exchange located in Islamic republic of Iran. On the other hand, the process of neural network training (especially
supervised training) can be considered as an optimization matter, through which network tries to determine
unknown parameters (weights) so that prediction error(difference between neural network output and actual output )
is minimized as possible (refer to Karay and Silva,2004; p.225). Therefore, in order to determining these parameters,
different optimization algorithms such as genetic algorithm, algorithms of ant colony, particle swarm algorithm, etc.
are applied. Most of optimization algorithms are inspired from natural processes, and in presenting them no attention
has been devoted to other aspects of human evolution. However, one of newest optimization algorithms, whose basis
has been motivated by socio-human processes, is imperialist competitive algorithm (ICA), in which colonization
process is considered as a stage of man’s socio-political evolution, and by its mathematical modeling an algorithm
was obtained which was universal in a short time and applied in various areas.

As mentioned above, back propagation was considered as some of shortcomings about learning algorithm that in
present study it is tried to utilize a new algorithm as imperialist competitive algorithm for training multi-layered
perception (MLP). As a result, present study would be an attempt to compare the efficiency of BPNN and MLP-ICA
network with the aim of bankruptcy prediction in Tehran’s stock exchanges. By presenting precise predictive
models for bankruptcy, firms would be aware about crisis occurrence at on hand and on the other hand, investors
would be helped to distinguish desired and undesired opportunities for investment and move towards expected
returns.

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
32
 SCIENCE ROAD JOURNAL
Other parts of this study are as follows: in part two, research literature relating with bankruptcy would be offered. In
part three, introduction of utilized models and methodology of the research would be explained in details. Inpart
four, the results of models would be offered and in last part, final conclusions of the study would be explained.

2. A literature review

In this part, common research literature between bankruptcy prediction and artificial neural networks would be
reviewed. The history of utilizing neural network in studies relating with bankruptcy prediction returns back to two
previous decades, when Adam and Sharda (1990) were used artificial neural networks for bankruptcy prediction for
the first time and found this model as a suitable method.

In a study by Udo (1993) upon predicting bankruptcy, a comparison between multi-layered perception network and
multiple regressions was done and obtained results indicated the better performance of multi-layered perception
network towards multiple regressions. Also in a study by Joe et al. (1997) similar results were obtained. By
examining Korean firms from 1991 to 1993, they compared discriminate analysis models, case-based forecasting
and multi-layered perception network. Their results showed better performance of multi-layered perception network
towards other statistical models.

Based on data of 220 firms and utilizing multi-layered perception neural networks, Jang and colleagues offered a
model for bankruptcy prediction. By adding current assets/ current liability to Altman’s (1968) proposed variables
and also by considering direct and indirect effect of inputs on used neural networks’ outputs, they examined
considered population. The results of this study show that total accuracy of neural networks’ prediction is higher
than logistic regression. In a similar comparison in Fletcher and Goss’s (1993) study between neural networks and
Logit model conducted for bankruptcy prediction, neural network had higher accuracy compared with its
competitors.

Also in 2005, it was Lee et al. (2005) who compared supervised and unsupervised learning algorithms in neural
network, in order to predicting Bankruptcy. They compared the back propagation algorithm as a representative for
supervised algorithms with the algorithm of Cohonen as a representative for unsupervised algorithms in neural
network. In this study, Altman’s (1968) variables have been applied. The number of utilized samples in present
study is 113 pairs of bankrupt and healthy companies, which were chose among South Korea’s stock exchanges.
They concluded that back propagation algorithm has better performance compared with Cohonen algorithm.

Tseng and Hu (2010) have compared 4 methods for bankruptcy prediction, which among utilized models the name
of MLP neural networks trained by back propagation algorithm and RBF networks is obvious. Among 4 utilized
models, RBF network has higher accuracy towards others.

3. Research methodology

The definition provided in this study for bankruptcy is the same definition in141 article of Iranian commercial law
for bankruptcy so that we regard firms with accumulated losses more than half of capital as a bankrupt company.
The procedure is as follows: after selecting population and sample, research data is extracted and divided into two
data sets of training and testing. By using first type of data, we would elaborate on neural network training and then
by applying remaining data we would test considered networks. Applied population in this study consists of firms
present in Tehran stock exchange during period of 2001-2009 inclusive in 141 article of Iranian commercial law for
bankruptcy. Therefore, those firms that had following conditions were used for neural network training:

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
33
 SCIENCE ROAD JOURNAL

1-Those firms inclusive in 141 article of Iranian commercial law, should have been accepted in Tehran stock
exchange at least since 1999 (because data of two years before bankruptcy would be used)

2-Those firms selected in this study as bankrupt were not financial brokerage companies.

3-They are included in trade act of 141 during period of 2001 till 2009 (they have accumulated losses greater than
half of capital)

Through examinations accomplished by researchers in present study, 80 firms during this period are included in the
act. Because the number of firms for neural network training is important, it has been tried to apply most of
available data for neural networks training as much as possible. However, the ratios of some firms had considerable
difference with other samples in the study and it would lead to decrease in performance of neural network, hence
some of bankrupt firms would be eliminated from the study. Consequently, the number of bankrupt firms in this
study reached to 70.

Also, in an industry that every bankrupt company is active, a healthy company, which has nearest amount of assets
to that company, was selected. It is done so because a neural network with training data from healthy firms is able to
distinguish between these firms and bankrupt companies. Healthy firms are selected from those firms that bankrupt
pairs were present at Tehran’s stock exchange. There was a problem with this matter that due to having small size of
industry in some of industries, selecting healthy pair for bankrupt firms was not feasible, hence it was tried to select
among healthy firms in upstream and downstream industry, and in the case of lack of these companies, a healthy
company with the same assets from a non-similar industry would be selected. It can be considered as one of
limitations in this study.

In next step, research data is divided into two sets of training and testing data sets. To do so, among research data,
some shall be selected randomly and assigned to these sets. Therefore, 20-80 percent of data was assigned to
training and testing respectively.

3.1. Research variables

Since financial ratios have long history in predicting bankruptcy, we tried to apply financial ratios for this aim and
also due to special limitations in Altman’s proposed variable (1968)(Coats and Fant, 1993; Lacher et al., 1995;
Odom and sharda, 1990; Rahimian et al., 1993; Sharda and Wilson, 1996; Wilson and Sharda, 1994), we considered
these variable as input variable. These ratios including;

(1) Working capital/total assets; (2) retained earnings/total assets; (3) earnings before interest and taxes/total assets;
(4) exchange value equity/book value of total debt; (5) sales/total assets.

Utilized data is financial ratios of two years before bankruptcy in bankrupt firms and selecting healthy pairs. In order
to training neural network using supervised algorithms, output data or target data is needed; thereby a neural
network can be familiar with the relationship between ratios of each healthy company and bankrupt one and also
their outputs. In this research as well as other researches which are used for classification, we segregate the outputs
of research with two different class definitions. 0 shows healthy firms and 1 shows bankrupt firm. Predicting
accuracy is calculate on the base of network output and real output. This error is occurring when we classify healthy
firm as bankrupt firm and vice versa. Final comparative review in this study will be applied with regard to BPNN
neural network and neural network- colonial rivalry in accurate classification rate of healthy and bankrupt firms.
(Accurate prediction percentage)

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
34
 SCIENCE ROAD JOURNAL

3-2- Multi-layer perception neural network (MLP)

A neural network is composed of uniting some neural neurons. Neuron is a processing unit that plays essential role
in functioning artificial neural network. Fig.1represents a neural neuron, which is composed of some key elements
such as weight, w, activation function, 𝑓, and bias, b. output of this neuron is calculated based on equation 3:

Equation 3 𝑎 =𝑓 𝑏+𝑤∙𝑝

Where p is input vector(r is the amount of input vector’s dimension), w is the row vector of weight, b is the amount
of bias, and𝑓,the activation function and a, the neuron’s output.

𝑝1
𝑤1
𝑝2 y1
x1
𝑤2 (𝑝i wi + b)
n 𝑓(𝑝𝑖 𝑤𝑖 + 𝑏)
a
x2
...

...
𝑤𝑟

...

...
𝑝𝑟
ym
xr
𝑏

1 Input Layer Hidden Output

Layer Layer
Fig.1. A neural neuron Fig. 2. A Multi-Layered Perception

Middle layer or layers process received data from input layer and give it to output layer. Output of this network is
obtained by applying equation 4:

Equation 4

𝑦 = 𝑓𝑜𝑢𝑡 𝑏𝑜𝑢𝑡 + 𝑓ℎ 𝑏ℎ + 𝑝𝑊ℎ 𝑊𝑜𝑢𝑡

In which pis input vector, ،𝑏ℎ and𝑏𝑜𝑢𝑡 are biases of hidden and output layers respectively, 𝑊ℎ and 𝑊𝑜𝑢𝑡 are weight
matrices of hidden and output layers respectively, and 𝑓ℎ ‫ 𝑡𝑢𝑜𝑓و‬are activation functions of neurons located in hidden
and output layers and 𝑦 output vector of network(Fig.2).

Activation function utilized in middle layers is tangent sigmoid function, which is calculated through equation 5.

2
Yi 
(1  exp( 2 X j ))  1
Equation 5

The performance indicator of back propagation algorithm is sum of squared error; in other words in this algorithm it
is tried to update weight matrices and bias vectors so that mean square error is minimum. Hence, one of error
indicators has been applied in this study called mean square error (MSE). It is regarded as criterion for selecting
optimum structure in the study. Equation 6 shows how to evaluate it.

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
35
 SCIENCE ROAD JOURNAL
1 𝑛
Equation 6 𝑀𝑆𝐸 = 𝑖=1 ( 𝑇 − 𝑂𝑢𝑡𝑁𝑒𝑡 )2
𝑛

One of the important parts of neural network is related to its training. Main difference of used models in this
research is neural network training algorithm and the first most applicable model for predicting bankruptcy is Back
propagation and the second one is colonial rivalry which are new algorithm. One of its manifest characteristics is the
ability to void from trap of local minimum.

3-2-1- imperialist competitive algorithm (ICA)

Imperialist competitive algorithm (ICA) is a new algorithm in the field of evolutionary computations and like other
evolutionary algorithms. It is a population-based stochastic search algorithm. It has been introduced by Atashpaz
and Lucas, recently (Rajabioun et al., 2008a, 2008b; Atashpaz-Gargari and Lucas, 2007a, 2007b; Atashpaz-Gargari
et al., 2008a, 2008b; Roshanaei et al., 2008; Jasour et al., 2008). Since then, it is used to solve some kinds of
optimization problem (Rajabioun et al., 2008a, 2008b; Atashpaz- Gargari and Lucas, 2007a, 2007b; Atashpaz-
Gargari et al., 2008a, 2008b; Roshanaei et al., 2008; Jasour et al., 2008). The algorithm is inspired by imperialistic
competition. It attempts to present the social policy of imperialisms to control more countries and use their sources
when colonies are dominated by some rules. If one empire loses its power, the rest of them will compete to take its
possession. In ICA, this process is simulated by individuals that are known as countries. This algorithm starts with a
randomly initial population and objective function which is computed for them. The most powerful countries are
selected as imperialists and the others are colonies of these imperialists .Then the competition between imperialists
take place to get more colonies .The best imperialist has more chance to possess more colonies. Then one imperialist
with its colonies makes an empire. Fig. 3 shows the initial populations of each empire (Rajabioun et al., 2008a,
2008b; Atashpaz-Gargari and Lucas, 2007a, 2007b; Atashpaz-Gargari et al., 2008a, 2008b; Roshanaei et al., 2008;
Jasour et al., 2008). If the empire is bigger, its colonies are greater and the weaker ones are less. In this figure
Imperialist 1 is the most powerful and has the greatest number of colonies.

Fig 3. Generating the initial empires (Rajabioun et al, 2008)

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
36
 SCIENCE ROAD JOURNAL

Fig.4. Moving colonies toward their related imperialist (Niknam et al, 2011)

After dividing colonies between imperialists, these colonies approach their related imperialist countries. Fig. 4
represents this movement. Based on this concept each colony moves toward the imperialist by a units and reaches its
new position. Where a is a random variable with uniform (or any proper) distribution, β, a number greater than1,
causes colonies move toward their imperialists from different direction and S is the distance between colony and
imperialist (Niknam et al, 2011).

a˷U(0, β×S)

It should be mentioned that every problem in optimization needs a cost function, which the process of optimization
would be implemented for its minimization. Respecting efficiency index in this network, the cost function is
determined based on the value of MSE. Along with each evolution circle (it is called decade in this algorithm) in
imperialist competitive algorithm, the best countries would be utilized as weights of neural network and for those
weights, the process of optimization would be continued until reaching the desired value of accuracy. However,
other stop conditions such as certain number of iterations can be applied.

4. Network testing and results analysis

After training research networks, it is the turn of offering test data to the research and recording relevant results.
Details related to the way of prediction in these models have been provided in table.1. According to table.1, MLP-
ICA neural network could have classified 85.71 % of healthy firms and 78.57 % of bankrupt ones properly. Also,
error type 1 for this model is 14.29% and error type 2 is 21.43%. Total results indicate that the performance of MLP-
ICA network was in such a way that 82.14% of firms present in test data have been classified properly.

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
37
 SCIENCE ROAD JOURNAL
Table.2 The results obtained from testing MLP and MLP-ICA networks

Model Sum of Status The number The number The The The
prediction of correct of correct errors percentage error
predictions predictions type 1 of correct type 1 or
and 2 predictions 2

28 healthy 14 12 2 85.71% 14.29%

MLP-ICA bankrupt 14 11 3 78.57% 21.43%

total 28 23 5 82.14% 17.86%

28 healthy 14 13 1 92.85% 6.15%

BPNN bankrupt 14 12 2 85.71% 14.29%

total 28 25 3 89.28 % 10.72%

Looking at performance of BPNN in table.2, it could be found that it has suitable power in predicting bankruptcy.
The details about prediction of this network indicate that its performance is very similar to the model of MLP-ICA.
Also, BPNN could show a 92.85% performance for predicting healthy firms with 13 correct predictions out of 14
predictions, in relation to predicting bankrupt firms we can call BPNN model performance better than its rival
model. (With 85.71 % accurate prediction)

According to table.2 the comparison of these two models of the study reveals that there is no significant difference
between the performance of MPL-ICA and BPNN in predicting healthy and bankrupt companies, However BPNN
network has one more accurate prediction than MLP-ICA (in both healthy and bankrupt firm) and also higher
performance.

5. Conclusion

Main aim of this study is identifying optimal neural network algorithm in order to predict bankruptcy. BPNN neural
network due to simplicity in training and efficiency in predicting bankruptcy is used. Also MLP network is used as
rivalry model which is integrated with one optimal algorithm called colonial rivalry. This algorithm is used for MLP
network training and in fact for determining weight of networks. Applied data in this study includes 70 bankrupt
firms inclusive in 141article of Iranian commercial law relating bankruptcy and 70 healthy firms in period of 2001-
2009 that are working in Tehran stock exchange.

We divided study data to training and examining and allocated 80 and 20 percent of all study data to each group.
With regard to accuracy of prediction as a measure for research model performance, it indicated that BPNN network
has better performance in compare with MLP-ICA model with 92.85 % of healthy firms and 85.71 % of bankrupt
firms. This is because MLP-ICA can classify 85.71% of healthy and 78.57 % of bankrupt firms with accuracy. As a
result BPNN has higher accuracy in compare with MLP-ICA, but this interval is trivial and insignificant.

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
38
 SCIENCE ROAD JOURNAL
Research limitation

Being small of population of accepted firms in Tehran’s stock exchange caused we could not pair non-bankrupt
firms with bankrupt ones in all of industrial aspects.

References

Altman, E. (1968). Financial ratios, discriminate analysis and the prediction of corporate bankruptcy, The Journal of
Finance, 23(4): 589-609.

Anandarajan, M, Lee, P, Anandarajan, A. (2004). Bankruptcy predication using neural networks, Business
Intelligence Techniques: A Perspective from Accounting and Finance, Germany: Springer-Verlag.

Atashpaz-Gargari, E., Lucas, C. (2007a); Imperialist competitive algorithm: an algorithm for optimization inspired
by imperialistic competition. In: Proceed- ings of the IEEE Congress on Evolutionary Computation, Singapore, pp.
4661–4667

Atashpaz-Gargari, E., Lucas, C. (2007b); Designing an optimal PID controller using Colonial Competitive
Algorithm; In: Proceedings of the First Iranian Joint Congress on Fuzzy and Intelligent Systems, Mashhad, Iran

Atashpaz-Gargari, E., Hashemzadeh, F., Lucas, C. (2008a); Designing MIMO PIID controller using colonial
competitive algorithm: applied to distillation column process. In: Proceeding of IEEE CEC 2008, within IEEE
WCCI 2008, Hong Kong, pp. 1929–1934.

Atashpaz-Gargari, E., Hashemzadeh, F., Rajabioun, R., Lucas, C. (2008b); Colonial competitive algorithm: a novel
approach for PID controller design in MIMO distillation column process; International Journal of Intelligent
Computing and Cybernetics (IJICC) 1 (3), 337–355

Beaver, W. (1966); Financial ratios as predictors of failure, Journal of Accounting Research, 5, 71-111.

Chuang, C. C., Jeng, J. T., Lin, P. T. (2004). Annealing robust radial basis function networks for function
approximation with outliers; Neuro computing, 56, 123–139

Coats, P.K, Fant, L.F. (1993); Recognizing financial distress patterns using a neural network tool, Financial
Management, 142-155

Dutta, S, Shekhar, S, Wong, W.Y. (1994). Decision support in non-conservative domains: Generalization with
neural networks, Decision Support Systems, 11(5), 527-544.

Fletcher, D, Goss, E. (1993). Forecasting with neural networks: An application using bankruptcy data, Information
&Management, 24, 159-167

Grudnitski, G. Osburn, L. (1993). Forecasting S and P and gold futures prices: An application of neural networks,
The Journal of Futures Markets, 13(16), 631-643.

Hagan, M. T. Dcmuth, H. B. Beale, M. (1996) Neural Networks design, PWS Publisher.

Hertz, J, Krogh, A., Palmer, R. G. (1991). Introduction to the theory of neural computation, Reading, MA: Addison-
Wesley.

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
39
 SCIENCE ROAD JOURNAL
Jang, J. S. R. (1993). ANFIS: Adaptive-network-based fuzzy inference systems. IEEE Transactions on Systems,
Man, and Cybernetics, 23(3), 665–685.

Jang, J. S. R, Sun, C. T, Mizutani, E. (1997). Neuro-fuzzy and soft computing: A computational approach to
learning and machine intelligence. NJ: Prentice-Hall.

Jasour, A.M., Atashpaz Gargari, E., Lucas, C. (2008); vehicle fuzzy controller design using imperialist competitive
algorithm; In: Proceedings of the Second First Iranian Joint Congress on Fuzzy and Intelligent Systems, Tehran,
Iran

Jensen, H.L. (1992) Using neural networks for credit scoring, Managerial Finance, 18(6), 15-26.

Joe, H. Y, Han, I. G, Lee, H. Y. (1997). Bankruptcy prediction using case-based reasoning, neural networks, and
discriminate analysis, Expert Systems with Applications, 13(2), 97–108.

Karray, F.O, Silva,C (2004). Soft Computing and Intelligent Systems Design, Pearson Education Limited

Kim, J.W, Weistroffer, H.R, Redmond, R.T. (1993). Expert systems for bond rating: A comparative analysis of
statistical, rule based and neural network systems, Expert Systems, 10(3), 167-172.

Kryzanowski, L. Galler, M. (1995).Analysis of small-business financial statements using neural nets, Journal of
Accounting, Auditing and Finance, 10(1), 147-172.

Lacher, R.C. Coats, P.K, Sharma, S.C, Fant, L.F. (1995); A neural network for classifying the Financial health of a
firm, European Journal of Operations Research, 85, 53-65.

Lee, K. (2001). Pattern classification and clustering algorithms with supervised and nsupervised neural networks in
financial applications, Ph.D. dissertation: Kent State University.

Lee, k, Booth, D, Alam, P. (2005); A comparison of supervised and unsupervised neural Networks in predicting
bankruptcy of Korean firms, Expert Systems with Applications, (29); 1-16.

Nam, M. D., Thanh, T. C. (2003). Approximation of function and its derivatives using radial basis function
networks, Applied Mathematical Modeling, 27(3), 197–221.

Niknam, Taher, Elahe Taherian Fard, Narges Pourjafarian, Alireza Rousta (2011). “An efficient hybrid
algorithm based on modified imperialist competitive algorithm and K-means for data clustering”. Engineering
Applications of Artificial Intelligence, 24, 306–317

Odom, M, Sharda, R, (1990). Bankruptcy prediction using neural networks, In Proceedings IEEE International
conference on neural networks, (pp. 133–168); San Diego

O’Leary, D. E. (1998).Using neural networks to predict corporate failure, International Journal of Intelligent
Systems in Accounting, finance and Management, 7, 187–197.

Pitts, W., and McCulloch, W.S. (1947) “How we know universals: the perception of auditory and visual forms,”
Bulletin Math. Biophys, 9, 27–47.

Piramuthu, S, Shaw, M.J, Gentry, J.A. (1994).A classification approach using multi-layered neural networks,
Decision Support Systems, 11(5), 509-525.

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
40
 SCIENCE ROAD JOURNAL
Rahimian, E, Singh, S. Thammachote, T, Virmani, R. (1993). Bankruptcy prediction by neural network; Neural
Networks in Finance and Investing: Using Artificial Intelligence to Improve Real-World Performance, R. Trippi and
E. Turban (eds.). Chicago, IL: Probus Publishing Co.

Sharda, R. Wilson, R.L. (1996). Neural network experiments in business-failure forecasting: Predictive performance
measurement issues, International Journal of Computational Intelligence and Organizations, 1 (2), 107-117.

Stam, A, Sun, M, Haines, M. (1996).Artificial neural network representations for hierarchical preference structures,
Computers and Operations Research, 23(12), 1191–1201.

Surendra, R., Krishnamurthy, A. (1997). Face recognition using transform features and neural networks. Neural
Networks, 30(10), 1615–1622.

Tam, K. Y, Kiang, M. Y. (1992); Managerial applications of neural networks: the case of bank failure predictions.
Management Science, 38(7), 926–947.

Tseng, F.M. Hu, Y.C. (2010). Comparing four bankruptcy prediction models: Logit, quadratic interval logit, neural
and fuzzy neural networks, Expert Systems with Applications, (37), 1846-1853

Udo, G. (1993); Neural network performance on the bankruptcy classification problem, Computers and Industrial
Engineering, 25(1) - 4, 377-380

Vythoulkas, P. C., Koutsopoulos, H. N. (2003). Modeling discrete choice behavior using concepts from fuzzy set
theory, approximate reasoning and neural networks, Transportation Research, 11, 51–73

Wilson, R.L, Shard, R. (1994) .Bankruptcy prediction using neural networks, Decision Support Systems,11 ،545-
557.

Wong, B.K, Selvi, Y. (1998).Neural network applications in finance: A review and analysis of literature (1990-
1996), Information and Management, (34), 129-139.

Yang, Z, Platt, M, Platt, H. (1999); Probabilistic neural networks in bankruptcy prediction‫و‬Journal of Business
Research 44: 67-74.

Zhang, G., Hu, M. Y, Patuwo, B. E, Indro, D. C. (1999); Artificial neural networks in bankruptcy prediction: general
framework and cross validation analysis. European Journal of Operational Research, 116, 16–32.

Corresponding Author: Hossein Parsian

Email: hparsean@yahoo.com
Acceptance Date: 9/19/2013
41