Bonfring International Journal of Man Machine Interface, Vol.

1, Special Issue, December 2011 10

ISSN 2250 1061 | 2011 Bonfring
Abstract--- The granting of loans by banks is one of the
key areas concerning decision problems that need subtle care.
During the last few decades, banks have relied on the personal
assessment of loan risks or on the traditional statistical
methods to predict the default of loans instead of using a
standardized evaluation tool. Despite the increase in
consumer loans defaults and competition in the banking
market, most of the Jordanian commercial banks are reluctant
to use artificial intelligence software systems for supporting
loan decisions. Application of data mining techniques for
predicting loan default assumes paramount importance in
Banking and financial services because of its performance.
This study developed a proposed model that identifies
artificial neural network as an enabling tool for evaluating
credit applications to support loan decisions in the Jordanian
Commercial banks. The analytics involved in this context pave
the way for evolving robust credit scoring models and
automation of the lending process. They also help discern the
pattern of relationship between the input [borrower
characteristics] and the output [loan default status]. This
paper uses Support Vector Machine which is a kind of neural
network for classification in order to grant the loan. The
experimental results indicate that the proposed technique
results in better prediction when compared to the conventional
techniques.

Keywords--- Artificial Neural Network, Data Mining,
Artificial Intelligence
I. INTRODUCTION
REDIT loans constitute a cornerstone of the banking
industry. The performance of loan contracts in good
standing guarantees profitability and stability of a bank.
Screening the customers financial history and financial
background is a very significant factor before any credit
decision is taken and it is a key process in reducing credit risk.
Loan approval will be given to good applicants with low credit
risk, whereas high risk applications will be rejected
(Limsombunchai et al., 2005). Therefore, banks should control
credit management thoroughly. Moreover, the undergoing
continuous changes in the business environment, the banking
and credit regulations, the marketing strategies of the banks,
the competition in banks lending strategies and the borrowing
patterns of customers, demand frequent revision and smart

P. Panneerselvan, Gobi Arts and Science College, Gobichettipalayam, E-
mail: p.panneerselvan@gmail.com
Dr.V. Thiagarasu, Associate Professor, Gobi Arts and Science College,
Gobichettipalayam.
responses. In addition, banking loan decisions require the use
of huge and various data and substantial processing time to be
able to serve a large number of variables and a variety of
different cases related to different customers. Thus, it will be
very hard to design or even to use a banking loan model that is
capable of tracing all possible and real responses and
interactions. Traditional models are unable to exhibit
uncertainty in the banking loan environment (Shachmurove,
2002). A set of creditscoring models that correctly classify
loan applications have been developed to support traditional
judgmental methods (Malhotra and Malhotra, 2003).
Furthermore, a challenge for todays commercial banks is their
ability to understand large amounts of information and reveal
useful knowledge to improve decision-making. Modern bank
managers are flooded in data. The sustainability of their banks
depends on their capabilities to sift through large volumes of
data, to extract useful knowledge and enforce this knowledge
in their decisions (Mitchell and Pavur, 2002). An intelligent
information system that is based on artificial intelligence will
provide managers with added value information, to reduce the
uncertainty of the decision outcome and to enhance banking
service quality. Thus, the application of new technologies can
give a bank a competitive advantage and lead to a higher
performance.
This study aims to develop a loan decision model using the
artificial neural network for the Jordanian commercial banks.
This is the first empirical research of its kind in our region that
addresses in a systematic way the issue of using artificial
neural networks in loan applications evaluations. Neural
networks with their capability of capturing nonlinear and
complex relationships are a powerful alternative to the
conventional forecasting and classification methods. Neural
networks are consistent paradigms of the nonparametric
approach in financial modeling due to their ability to correctly
classify and predict consumer loan defaults. The purpose of
using the neural network algorithm in bank loan decisions is to
simplify a loan officers job, to control it and to achieve more
efficiency and productivity (Curry and Moutinho, 1993).
Neural networks are used because of the following reasons:
Neural networks promise to be a breakthrough in areas
where a traditional computer system and statistical
methods have difficulty to support the decision making
process in todays complex business environment. Neural
networks have been used in building intelligent
information systems that mimic the way in which humans
think. They are suitable for modeling business logical
outcomes where variables give unstable signs and interact
in a complex non-linear way (Peel and Wilson, 1996)
Loan Decisions based on Extreme Learning
Machine
P. Panneerselvan and Dr.V. Thiagarasu


C

Bonfring International Journal of Man Machine Interface, Vol. 1, Special Issue, December 2011 11
ISSN 2250 1061 | 2011 Bonfring
Artificial neural networks are promising tools for
modeling noisy, inconsistent or incomplete data in
different business situations. The application software is
self-adjustable for any user and it only requires a
representative training set against a standard reference.
Then, the ANNs self-adaptive nature will take over
(Pollard, 1990)
Most financial decisions have incomplete information, so
neural networks have more applications for such data than
do expert systems and advanced statistical methods (Tafti
and Nikbakht, 1993)
The advantages of using ANN are as follows:
Eliminates the necessity of having a linear relationship
between input variables and the output. The network
simply learns the relationships based on the representative
sample data and produces output (the relationship can be
linear or nonlinear). This reduces the risk of bias in
decision making that might occur if an incorrect
functional form or relationship is assumed.
Enables the receipt of output even when presented with
input that it has never seen before.
Has the ability to learn from past experience and provide
a better output as the system is trained with new
examples. This also reduces subjectivity in decision-
making. However, the main problem with using the ANN
is that it does not tell the system user how it arrived at the
results. These obviously depend on the correctness and
completeness of the data that has been entered. The
system is good at processing and recognizing patterns, but
it may be exposed to the risk of producing something out
of nothing. Therefore, the data used for training must
represent a universal set of parameters.
Extreme Learning Machines is used in this paper which is
a Single Hidden Layer Feed-Forward Neural Networks. It has
very high capability that can resolve problems of data
regression and classification. Certain challenging constraints
on the use of feed-forward neural networks and other
computational intelligence approaches can be overcome by
ELM. Due to the growth and improvement in the ELM
techniques, it integrates the advantages of both neural
networks and support vector machines by having faster
learning speed, requiring less human intervene and robust
property.
II. RELATED WORKS
Artificial neural networks have been fruitfully used in a
variety of business fields including marketing, accounting,
management information systems and production management
(Cao and Parry, 2009). Most of the studies have used neural
networks for predicting future stock behavior, financial crises,
bankruptcy, exchange rate and detecting credit card fraud.
Furthermore, researchers have used neural networks in
modeling market response collective behavior,
telecommunication and real estate evaluations. ANNs are also
used for analyzing relations among economic and financial
phenomena, forecasting, generating time series, optimization
and decision making (Shachmurove, 2002). Artificial neural
networks have successfully provided effective credit
evaluations for supporting granting loans. The Security Pacific
Bank was the first bank to announce neural networking system
to the public in 1991 (Goonatilake and Treleaven, 1995).
Researchers are currently focusing on using neural network
classification models and particularly backpropagation neural
networks in classifying loan applications into good and bad
ones.
Customer loans include short and intermediate term loans
that are extended through regular business channels to finance
the purchase of goods and services for individual consumption
(Malhotra and Malhotra, 2003). This research attempts to
explore whether using neural networks will provide more
accurate personal loan decisions in the Jordanian commercial
banks. It will also propose a neural computing model as the
basis for a decision support tool in granting or rejecting bank
loans. The model will apply banking credit standards to
determine the customer who will be eligible for credit
approval. A customer will be qualified for credit after meeting
the acceptable standards set by the banking institution.
The major risk that a bank faces is the probability of a
customers default. An experienced credit manager should be
able to make the right judgment of the probability of default
for different types of customers. To have effective managerial
control over the credit department requires fast, accurate and
real-time information that must be processed in a fair and
objective manner. Artificial intelligence technologies have
proven to be a fruitful investment in the banking industry,
such as neural networks, for their ability to assist in improving
the quality of credit decisions to reduce credit risk
(Limsombunchai et al., 2005). Thus, artificial neural networks
with their capabilities of capturing nonlinear and complex
relationships are considered promising techniques in
classification problems (Zhang, 2004; Huang et al., 2004).
Traditional judgmental methods and conventional
statistical techniques are used in the approval of loan
worthiness (Tafti and Nikbakht, 1993). Traditional statistical
models such as discriminant analysis and logistic regression
usually assume multivariate normality and homoscedastic and
these assumptions are often violated in the real world banking
data (Giang, 2005; Huang et al., 2004). Moreover, traditional
techniques lack the reasoning skills that an experienced loan
officer possesses. The current research views the neural
network as superior alternatives to the traditional techniques
that are used in loan application evaluations.
III. METHODOLOGY
A. Extreme Learning Machine
ELMs parameters can be analytically determined rather
than being tuned. This algorithm provides good generalization
performance at very fast learning speed. From function
approximation point of view ELM is very different compared
to the traditional methods. ELM shows that the hidden node
parameters can be completely independent from the training
data.
In conventional learning theory, the hidden node
parameters cannot be created without seeing the training
data.

Bonfring International Journal of Man Machine Interface, Vol. 1, Special Issue, December 2011 12
ISSN 2250 1061 | 2011 Bonfring
In ELM, the hidden node parameters can be generated
before seeing the training data.
B. Salient Features of ELM
Compared to popular Back propagation (BP) Algorithm
and Support Vector Machine (SVM), ELM has several salient
features:
Ease of use: Except predefined network architecture, no
other parameters need to be manually tuned. Users need
not have spent much time in tuning and training learning
machines.
Faster learning speed: The time taken for most of the
training will be in milliseconds, seconds, and minutes.
Other conventional methods cannot provide such a fast
learning speed.
Higher generalization performance: The generalization
performance of ELM is better than SVM and back
propagation in most cases.
Applicable for all nonlinear activation functions:
Discontinuous, differential, non-differential functions can
be used as activation functions in ELM.
Applicable for fully complex activation functions:
Complex functions can also be used as activation
functions in ELM.
Extreme Learning Machine (ELM) meant for Single
Hidden Layer Feed-Forward Neural Networks (SLFNs) will
randomly select the input weights and analytically determines
the output weights of SLFNs. This algorithm tends to afford
the best generalization performance at extremely fast learning
speed.
The structure of ELM network is shown in figure 1. ELM
contains an input layer, hidden layer and an output layer. The
ELM has several interesting and significant features different
from traditional popular learning algorithms for feed forward
neural networks. These include the following:

Output
Layer

First
Hidden
Layer
Input Layer
Figure 1: Structure of ELM Network
The learning speed of ELM is very quick when compared
to other classifier. The learning process of ELM can be
performed in seconds or less than seconds for several
applications. In all the previous existing learning
techniques, the learning performed by feed-forward
network will take huge time even for simple applications.
The ELM has enhanced generalization result when
compared to the gradient-based learning techniques. The
existing gradient-based learning technique and a few
other learning techniques may encounter several problems
like local minima, not proper learning rate and over
fitting, etc. To overcome these problems, some techniques
like weight decay and early stopping techniques must be
utilized in these existing learning techniques.
The ELM will attain the results directly without such
difficulties. The ELM learning algorithm is much simple
than the other learning techniques for feed-forward neural
networks. The existing learning techniques can be applied
to only differentiable activation functions, whereas the
ELM learning algorithm can also be used to train SLFNs
with many non-differentiable activation functions.
C. Extreme Learning Machine Training Algorithm
If there are N samples (x
i
, t
i
), where x
i
= [x
i1
, x
i2
x
in
]
T

R
n
and t
i
= [t
i1
, t
i2
, , t
im
]
T
R
n
, then the standard SLFN
with N hidden neurons and activation function g(x) is defined
as:

=1

= 0

, = 1, . , .,
where w
i
= [w
i1
, w
i2
, , w
in
]
T
represents the weight vector
that links the ith hidden neuron and the input neurons,
i
= [
i1
,

i2
, ,
im
]
T
represents weight vector that links the ith neuron
and the output neurons, and b
i
represents the threshold of the
ith hidden neuron. The . in w
i
. x
j
indicates the inner product
of w
i
and x
j
. The SLFN try to reduce the difference between o
j

and t
j
. This can be given as:

=1

, = 1, . , .,
or, more in a matrix format as H = T, where

1
, . ,

, ,

,
1
, . ,

1
,
1
+
1

1
,
1
+
1

and =

The matrix H is the hidden layer output matrix of the
neural network. If the number of neurons in the hidden layer is
equal to the number of samples, then H is square and
invertible. Otherwise, the system of equations requires to be
solved by numerical methods, concretely by solving

| |

Bonfring International Journal of Man Machine Interface, Vol. 1, Special Issue, December 2011 13
ISSN 2250 1061 | 2011 Bonfring
The result that reduces the norm of this least squares equation
is

=
where H

is known as Moore-Penrose generalized inverse. The

most significant properties of this result are:
Minimum training error.
Smallest norm of weights and best generalization
performance.
The minimum norm least-square solution of H = T
is unique, and is

=
The ELM algorithm works as follows
Give a training se = (
1,

1

)|
1

,
1

, 1 =
1.

t activation function g(x) and hidden neuron

, do
the following
Assigning random value to the input weight

and the
bias

, = 1, .

Find the hidden layer output matrix H.
Find the output weight , using =HT, where , H and
T are defined in the same way they were defined in the
SLFN specification above.
IV. CONCLUSION
Although the credit decision approval in the Jordanian
banks has been subjective and it is up to the loan officer in
most cases, banks can improve their credit analysis methods
through utilizing artificial intelligence technologies such as
neural networks. This study has proposed a new technique to
evaluate loan application as a decision support system in the
field of banking credit system; the proposed neural network
model was designed to represent the most influential factors to
the credit decision in Jordanian Banks. These factors were
taken from the guidelines that loan officers use in the
Jordanian banks. Besides, banks can customize the system
according to their specific strategies. This paper uses Extreme
Learning Machine (ELM) meant for Single Hidden Layer
Feed-Forward Neural Networks. This proposed technique has
better ability for classification which is suggested in the
experimental results. Furthermore, the high accuracy of this
model proves that the proposed technique can be useful for
classifying loan applications.
REFERENCES
[1] Duch W. Swaminathan K. And Meller J., (2007). Artificial Intelligence
Approaches for Rational Drug Design and Discovery, Current
Pharmaceutical Design, 2007, 13.
[2] Brachman, Ronald J, (2006). (AA)AI More Than the Sum of Its Parts,
AI Magazine Volume 27 Number 4.
[3] Cassimatis, N. Mueller, E. And Winston P. (2006). Achieving Human-
Level Intelligence through Integrated Systems and Research, AI
Magazine Volume 27 Number 2 AAAI.
[4] Shorouq Fathi Eletter, Saad Ghaleb Yaseen and Ghaleb Awad Elrefae,
Neuro-Based Artificial Intelligence Model for Loan Decisions,
American Journal of Economics and Business Administration 2 (1):
27-34, Science Publications 2011.
[5] Zurada, J, Could Decision Trees Improve the Classification Accuracy
and Interpretability of Loan Granting Decisions, System Sciences
(HICSS), 2010 43rd Hawaii International Conference, Pp 1 - 9
[6] Hongxun Jiang; Xiang Ye; Study on Loan Auditing Decision Model
Based on Plural Entropy, Intelligent Control and Automation, 2006.
WCICA 2006. The Sixth World Congress, vol 2, Pp 6925 6928
[7] Guo Jian; Debt financing, loan market structure and enterprise
investment decisions based on real option, Management Science and
Engineering, 2009. ICMSE 2009. International Conference, Pp 1768 -
1773
[8] Luo JianHua and Shen Tao Credit Risk Evaluation of Micro-loan
Companies on Soft Set, , 2010 International Conference on E-
Business and E-Government (ICEE), Pp 1110 1115.
[9] Fei Wang and Zhigang Song, Research on the Credit Risk Evaluation
and Forecast of Housing Mortgage Loans,. International Symposium
on Intelligent Information Technology Application Workshops, 2008.
IITAW '08.
[10] Shen Tao, Lending relationship analysis of micro-loan company on
game theory, , 2011 International Conference on Management
Science and Industrial Engineering (MSIE), Pp 149 152
[11] Li Yuanyuan, Research on Personal Credit Evaluation System of
Commercial Banks, First International Workshop on Knowledge
Discovery and Data Mining, 2008. WKDD 2008, Pp 457 460.
[12] Zurada, J, Could Decision Trees Improve the Classification Accuracy
and Interpretability of Loan Granting Decisions, 2010 43rd Hawaii
International Conference on System Sciences (HICSS), Pp 1 9.
[13] Hsu, C.F and Hung, H.F. Classification Methods of Credit Rating - A
Comparative Analysis on SVM, MDA and RST, International
Conference on Computational Intelligence and Software Engineering,
2009. CiSE 2009, Pp 1 4.
[14] Tsakonas, A., Ampazis, N. and Dounias, G. Towards a
Comprehensible and Accurate Credit Management Model: Application
of Four Computational Intelligence Methodologies, 2006
International Symposium on Evolving Fuzzy Systems.
[15] Shen Tao, Lending relationship analysis of micro-loan company on
game theory, 2011 International Conference on Management Science
and Industrial Engineering (MSIE), Pp 149 152
[16] Luo JianHua; Shen Tao; Credit Risk Evaluation of Micro-loan
Companies on Soft Set, 2010 International Conference on E-Business
and E-Government (ICEE), Pp 1110 1115
[17] Changling Yin, Fuzzy pattern recognition model of bank loan risk and
its application, 2010 Seventh International Conference on Fuzzy
Systems and Knowledge Discovery (FSKD) Pp 1272 1276
[18] Fangyuan Lu and Min Lu Evolutionary Game Analysis on the
Phenomenon of National Student Loan Default, International Seminar
on Business and Information Management, 2008. ISBIM '08.
[19] Workineh, A. and Homaifar, A. Robust bidding in LCS using loan
and bid history, Aerospace Conference, 2010 IEEE Pp 1 8
[20] Lixin Yu and Yan-Qing Zhang Evolutionary Fuzzy Neural Networks
for Hybrid Financial Prediction, IEEE VOL. 35, NO. 2, MAY 2005
[21] G. Winter, J. Periaux, M. Galan, and P. Cuesta, Genetic Algorithms in
Engineering and Computer Science. New York: Wiley, 1995.
[22] Yuanyuan Zhang, Shouyang Wang and Gengzhong Feng Loan to
Value to Newsvendor Based on Stock Document Mortgage, 2007
IEEE, Pp 3939-3942.
[23] Zunxin wu, Tranche Technology of Loan Securitization, 2007 IEEE
Pp 4113-4115.
[24] Feyza Merve Isik, Bulent Tastan and Pnar Yolum Automatic
Adaptation of BPEL Processes Using Semantic Rules: Design and
Development of a Loan Approval System, 2007 IEEE, PP 944-948

Bonfring International Journal of Man Machine Interface, Vol. 1, Special Issue, December 2011 14
ISSN 2250 1061 | 2011 Bonfring
[25] Wang Xiu-hua Liang Ling The Study on the Loan Risk Pricing of the
Bank in Internal Rating-Based Approach, 2008 International
Conference on Risk Management & Engineering Management,
IEEE2008, Pp 398-402.
[26] Yixue Li, On Loan-to-value Ratio of Inventory Financing under Static
Pledge and Dynamic Pledge Fashion, 2008 IEEE Pp 441-446