Download Supporting Educational Loan Decision Making Using

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Mathematical model wikipedia , lookup

History of artificial intelligence wikipedia , lookup

Pattern recognition wikipedia , lookup

Catastrophic interference wikipedia , lookup

Time series wikipedia , lookup

Convolutional neural network wikipedia , lookup

Transcript
Supporting Educational Loan Decision Making Using Neural Network
Nooraini Yusoff and Fadzilah Siraj
Faculty of IT
Universiti Utara Malaysia, 06010 Sintok
Kedah, Malaysia
Tel: +604-9284629, Fax: +604-9284753, E-mail: [email protected], [email protected]
ABSTRACT
still practicing manual or semi-manual processing. In
addition, due to inefficient filing system, some funding
institutions face a problem in tracing back the higher
learning leavers for payment back purpose (Siraj &
Yusoff, 2005).
This study introduces i-Neuro, a decision support
system that can assist in loan decision making by
educational loan funding institutions. i-Neuro is a
predictive system that integrates Neural Network
technique, thus can help the management to predict
which application to accept or reject. The prediction
can be done as Neural Network has trained previous
batch of loan application data and stored association
between application characteristics (attributes) that
explains which applications were accepted and rejected.
The association in previous data can predict the new
current application. The use of i-Neuro in loan
application processing reduces the management
workload by providing the list of eligible applicants
based on the merit agreed by the management. As a
result, the time required for loan application processing
can be reduced as well as allowing the management to
update the information in database easily. i-Neuro
system has been tested on real data. It has shown to
achieve satisfactory results and indicates its potential in
facilitating the loan application processing.
In improving routine office tasks that includes
application processing, data processing and reporting, a
number of several funding institutions have adopted
various versions of loan automated processing systems
to reduce the time consumed to gather, store and process
the applications (Siraj & Yusoff, 2004). This is
required due to increasing number of applications from
year to year.
Nevertheless, the jobs’ demands are not only for
documenting and reporting, most of the loan mortgaging
processes tasks deal with decision-making which some
decisions that may require the presence of experts. For
the purpose of aiding the loan decision-making, a
number of decision support systems have been
developed for specific funding instutions needs for
example Student Information System (SIS) and
Accounting System (Pusat Komputer UUM, 2004).
These systems are used to assist in decision making that
provide statistics, data analysis and graphical charts.
Keywords
Neural Network, Decision Support System, Educational
Loan, Intelligent System
For the analysis purposes, the decision support systems
are useful in summarizing data, processing data to
information and representing diagnostics information.
Nevertheless, it lack of prediction capability which
some institutions might need this facilities to answer
questions such
as “what is the characteristic of an application to be
accepted?”, “what is other factors to be considered other
that the academic qualifications?” and “what is the loan
quota-university practiced?” The answers can only be
provided through a series of analysis and decision
making meeting which attended by the experts who
have years of experience to predict and explain certain
occurrences.
1.0 INTRODUCTION
In supporting the learning of higher education in
Malaysia with a well conducive learning environment,
loan or scholarship are funded to students for their
allowances and fees . For this purpose, loans or
scholarships are granted by a number of institutions and
organizations that includes PTPTN, JPA, MARA and
state institutional bodies.
From a managerial point of view, a few processes and
decision makings are required for ensuring loan or
scholarship are granted to an eligible student with a
sufficient of amount. The funding institution usually
employs loan officers to make credit decisions or
recommendations for that particular institution. These
officers are given some hard roles in evaluating the
worthiness of each application. As the number of
applications increases every year, the loan processing
and approval tasks have become more challenging and
difficult especially for the funding institutions which are
Hence, the needs to mimic or replicate human thinking
processes have motivated the research in Artificial
Intelligence (AI). The capability of AI techniques in
predicting, diagnosing and advising has made AI
becoming popular in decision-making. A number of AI
techniques have been integrated into decision support
system software such as SAS, Clementine and Data
Mining software.
422
of applicants increases year by year, the application
processing has become challenging and difficult, which
creates problems when most of the processes are still
done manually with no systematic filing system and
database. Instead of using hardcopy form, a few
managements have moved to online form for the
application purposes (Lembaga Biasiswa Negeri Kedah,
2002; JPA, 2005). This may reduce the task of
encoding and storing applicants’ information into a
database. Nevertheless, evaluation to select or shortlist
applicants are still consuming more time. The approval
of any loan or scholarship is time-consuming as the
evaluation is done manually based on human
judgement. Recently, a few issues arise due to qualified
student were not granted applied loan or scholarship.
This could be due to flaws in judging the loan or
scholarship. In addition, inefficient file system also
could create problems in tracing the student’s status for
the payment back purposes. This issue has also been
raised in our media that reported the funding institutions
are facing problems regarding loan payment back by the
graduated students.
Hence, this study introduces Neural Network (NN)
system, one of the most important branches of Artificial
Intelligence that is designed with forecasting capability.
Researchers have begun to assess the usefulness of NN
methodologies beyond predictive accuracy toward
developing a deeper understanding of trends and
patterns (Hardgrave, 1994; Hammond, 2000). NN can
provide advantages in projecting and understanding the
future education spending (Hanushek, 2001). NN
requires a set of data, and from this data, the association
between variables in the data are obtained to predict or
forecast new data (Teng, 1996). Therefore, for the
problem in hand, NN can be used to predict a loan
application whether can be accepted or rejected.
In this study, a decision support system that integrates
NN techniques known as i-Neuro has be developed for
improving loan application processing that such
integration can be beneficial to assist the decision maker
in making decisions regarding loan application (Siraj &
Yusoff, 2005).
.
2.0
Hence, in this study, an integration of office automation
system and intelligent decision support systems is
proposed. Web based automation system can assist
applicants and management in performing loan
application processing, while intelligent decision
support system that integrates Neural Network, can be
as a helpful tool in loan approval decision making.
EDUCATIONAL LOAN DECISION
MAKING
In general, application for loan or scholarship is
announced in every academic term each year. A funding
institution announces the release of the application form
via newspaper and or media, but there are still
institutions that require applicants to go to the main
office to get and to return the application form
personally with an amount of price.
3.0
NEURAL NETWORK
With the advent of modern comp uter technology and
information science, sophisticated information systems
can be built that can make decisions or predictions
based on information contained in available past data.
Such systems are called learning systems and are
currently used for the purpose of classification and
prediction (Principe et al., 2000). Neural Networks are
popular techniques for classification and prediction
problem.
Starting from the submission, funding institution staffs
will collect all the application forms and calculate the
merit for each complete application form. Merit is given
by accumulating the total mark for each applicant based
on the criteria defined the institution. The application
process will take about couple of months, starting from
the submission of application up to listing out those
approved applicants.
A neural network system is an artificial intelligence
model that replicates the human brain’s learning
process. Tsoukalas & Uhrig (1997) define a neural
network as: “A data processing system consisting of a
large number of simple, highly interconnected
processing elements (artificial neurons) in an
architecture inspired by the structure of the cerebral
cortex of the brain.”
The range of merit that indicates the eligibility of the
applicants is determined by each institution. The list of
eligible applicants will then be tabled in the institution
committee board meeting, which is also held once a
year. In the meeting, the number of applicant selected is
based on the quota allocated for each public and private
university. For the number of applicant that is less than
the university’s quota, the applicant will be selected
based on the rank of merit. Once the approved
applicants are selected, the offer letters and the
agreement will then be posted to them. When the
acceptance letters are received from the applicants, their
profiles will be stored in files for further use (Siraj &
Yusoff, 2005).
In an artificial neural network, a number of inputs, or
attributes, and their corresponding outputs, or classes,
are given. A training algorithm uses these sample
inputs, called the training set, to design a decision
function that can accurately predict the class for any
sample thereafter. The algorithm response is compared
to the actual response to determine how well the
classifier performs (Barker et al., 2004).
In educational loan decision makings, every application
of loan needs to be processed and revised for every
criteria defined by particular funding bodies. As number
423
Nodes are used to represent the brain’s neurons and
these nodes are connected to each other in layers of
processing. Fig. 1 illustrates the three types of layers of
nodes: the input layer, the hidden layer or layers
(representing the synapses) and the output layer. The
input layer contains data from the measures of
explanatory or independent variables. This data is
passed through the nodes of the hidden layer(s) to the
output layer, which represents the dependent variable(s).
A nonlinear transfer function assigns weights to the
information as it passes through the hidden layer nodes,
mimicking the transformation of information as it
passes through the brain’s synapses. The goal of the
artificial neural network model is that the effect of these
weights will result in a response that is equivalent to the
response that would result from the relationship that
really exists between the input independent variables
and the output, or dependent, variable(s).
an applicant will be granted a loan or not based on the
merit determined by the management. At the same
time, the stored information can als o be accessed easily
whenever information is required by the management.
As Neural Network (NN) can be trained based on
previous loan application records, the forecasting model
for loan application can be established. The established
NN model can be applied to current or new application
form in order to determine the status of the applications.
Loan Officer
Predicted approvable
applicants
Information
Reporting Software
Reporting Software
Information
data
Prediction
results
NN
RESULT
data
Applicant’s Database
Information
Application Form
Pre-Processing
Pre
-
Applicant with
Applicant with
sufficient merit
Internet
Applicants
Figure 2: Overview of i-Neuro system
The NN approach involves 3 main processes that
include data preparation that preprocesses and cleanses
previous data to be trained and tested, establishing
prediction model that extracts a set of weight from the
trained and tested data and predicting education loan
that predicts result based on the extracted knowledge (a
set of weight).
Figure 1: Neural Network Architecture
In contrast, a traditional rule-based system would have
rules encoded within it that a designer has previously
identified. The advantage of neural network systems is
that it is not always possible for a human designer to
express and encode rules in a reasonable time-frame or
even express then at all. A further disadvantage of rulebased systems is that if the rules change for some reason
then it is necessary for the designer to reincorporate the
new rules within the rule-base (Barker et al., 2004). A
number of Neural Networks applications found in
literature include business (Hanke & Reitsch, 1998;
Hall, 2000; Siraj et al., 2003), economic (Siraj & Junoh,
2002), financial (Kaastra & Boyd,1996; Perez,1999;
Surkan, 1999; Siraj et al., 2000; Jingtao & Chew, 2001),
and education (Whitson, 1999; Carlson,2000;
Mullier,2001; Gonzalez & Des Jardins, 2002; Siraj &
Asman, 2002; Siraj & Rahman, 2003; Siraj & Sudin,
2003; Barker et al., 2004).
4.1 Data Preparation
The sampling consists of 1062 records of loan
application in the year 2002 from Lembaga Bisasiwa
Negeri Kedah (LBNK), one of the state educational
loan bodies that have been accepted and rejected. Many
of the questions found on the survey closely resemble
student attributes appearing in the LBNK’s loan
application form and other educational research.
The data set consists of respondents’ academic,
demographic, and socio-economic backgrounds have
been identified as the number of input nodes in the input
layer. The target output or the classification variable is
divided into two categories of loan application status,
reject or accept. The collected data was pre-processed
and cleansed prior to establishing the prediction model.
4.0 I-NEURO : EDUCATIONAL LOAN
DECISION SUPPORT SYSTEM
In improving the capability of decision support system
and loan processing, a predictive decision support
system, i-Neuro, has been developed. i-Neuro is a webbased intelligent decision support system that is used to
facilitate loan application processing and decision
making (Fig. 2). The loan application form can be filled
up and submitted online.
For each submitted
application, the information will be automatically stored
into the system’s database prior to determining whether
4.2 Establishing Prediction Model
To construct a prediction model, the data set of target
respondents requires training and testing processes.
Both processes are involved in obtaining a set of weight
coefficients ([v] and [w], see Fig. 3) that can be used in
predicting. For these purposes the data set was
424
partitioned into two groups; training (80%) and testing
(20%).
system capable to
application status.
[vij]
Qualification
z1
[wjk]
y1
Diploma
x2
z3
xi
y2
……….
zj
z_ in j = ∑ xivij
loan
Once the prediction model has been constructed, the
web based prediction module is able to predict the status
of an application (accept or reject) using a feedforward
Neural Network algorithm. The algorithm uses the
weight obtained from the training and testing processes
and also the Binary Sigmoid Function (as 1) to predict
the educational loan application status.
Matriculation
z2
…….
educational
Output layer
x1
Demographic
the
4.3 Predicting Educational Loan
Hidden layer
Input layer
predict
y_ in k = ∑ zjw jk
Figure 3: NN Architecture
application status, y k = 1/(1 + exp -y_in k)
The aim of training and testing was to obtain a
prediction model by varying the NN parameters that
include number of hidden units, the learning rate, the
momentum and the training stopping criteria. The
combination of these parameters that gave the optimal
NN performance indicated the best value of those
parameters to be included into the prediction model.
Fig. 4 depicts the parameter setting interface.
(1)
Prior to predicting the application status, the user of the
prediction system has to respond to some questions
based on his/her academic, demographic, and socioeconomic background. Fig. 5 question form and the
prediction result interface.
Figure 5: Question form and prediction result
Figure 4: Parameter setting interface
5.0 SUMMARY AND CONCLUSIONS
During the testing phase, 200 hundreds of applications
can be processed and predicted the acceptance status in
less than half an hour. The result also indicates that NN
obtained 99.06% prediction accuracy. Table 1 details
the prediction model constructed by the NN.
Neural Network technique embedded into i-Neuro can
help the management to predict which application to
accept or reject. This can be done as Neural Network
has trained previous batch of loan application data and
stored association between application characteristics
(attributes) that explains which applications were
accepted and rejected. The association in previous data
can predict the new current application data with same
characteristic but no loan acceptance result
Table 1: Neural Network model for predicting educational
loan
Parameter
Model
No. of input units
No. of hidden units
No.
of
output
classes
Learning rate
Momentum rate
Activation function
Stopping criteria
Value
Multilayer perceptron with
Backpropagation Algorthim
12
5
1
i-Neuro system can assist the management in evaluating
application. The management only needs to agree or
disagree to the system recommendations without
reviewing each application one by one. The use of iNeuro in loan application processing reduces the
management workload by providing the list of eligible
applicants based on the merit agreed by the
management. As a result, the time required for loan
application processing can be reduced as well as
allowing the management to update the information in
database easily. i-Neuro system has been tested on real
data. It has shown to achieve satisfactory results and
indicates its potential in facilitating the loan application
0.3
0.1
Sigmoid
500
Once a prediction model and a set of weights obtained,
they could be applied into a prediction system thus the
425
processing. In essence, the potential use of i-Neuro can
be accelerated to promote any organizations as an
efficient and effective organization that has competitive
advantage.
Lembaga Biasiswa Negeri Kedah (2002). “Permohonan
Pinjaman.”
[online]
<http://www.kedah.gov.my/lbnkweb/permohona
n.html >.
Mullier, D., Moore, D., & Hobbs, D. (2001). A NeuralNetwork System for Automatically Assessing
Students. World Conference on Educational
Multimedia,
Hypermedia
and
Telecommunications 2001(1), pp.1366-1371.
Perez, M. (1999). Neural Networks Applications in
Bankruptcy Forecasting: A State of the Art.
European Symposium on Intelligent Techniques.
Orthodox Academy of Crete, Greece
Principe, J.C., Euliano, N.R. & Lefebvre, W.C. (2000).
Neural and Adaptive Systems: Fundamentals
Through Simulations. John Wiley & Sons, Inc.
Pusat Komputer Universiti Utara Malaysia. (2004).
Laporan Penilaian Anugerah Perdana Teknologi
Maklumat 2004.
Siraj, F, & Asman, H (2002). Predicting Information
Technology
Competency
Using
Neural
Networks. Proceedings of the 7th Asia Pacific
Decision
Sciences
Institute
Conference.
Bangkok, Thailand. 24 – 27 July, pp. 249 – 255.
Siraj, F & Junoh, M. Z. M. (2002). Predicting the GDP
of the New Economy Based on the Human
Capital Using Neural Network Approach.
Proceedings of International Conference on
Artificial
Intelligence
and
Engineering
Technology. Kota Kinabalu, Sabah. 17 – 18 June.
Pp. 612 - 616
Siraj, F & Rahman, N A (2003). Building Academic
Achievement Prediction Model Using Neural
Networks,
Proceedings of International
Seminar on Learning and Motivation ISLM 2003,
Penang, Malaysia, 13 – 15 October
Siraj, F & Sudin, N. (2003). Determining A Suitable
Program for SPM Holders Using PRO-Predictor.
Malaysia – Japan Seminar on Artificial
Siraj, F., Zakaria, A., Ab. Aziz, A. & Abas, Z., (2003) A
Web Based Business Insolvency Classifier using
Neural Network. Proceeding of AIAI 2003.
Siraj, F, Zakaria, A., Yassin, A & Ishak, W. H .W.
(2000). Neural Networks Approch on On-line
Handwritten Signature Verification System.
Simposium Kebangsaan Teknologi Maklumat:
ITSym'2K 11-12 April. pp. 29-37. Universiti
Kebangsaan Malaysia.
Surkan, A. J. (1999). Pretesting and Data Modeling for
Forecasting Student Success in Computer
Science Courses. Proceedings of ASEE/IEEE
Frontiers in Education.
Teng, C L. & Lee, C.S. G. (1996). Neural Fuzzy
Systems. Prentice Hall
Tsoukalas, L.H. & R.E. Uhrig. (1997). Fuzzy and
Neural Approaches in Engineering. New York:
Wiley and Sons.
Whitson, (1999). An application of Artificial
Intelligence to Distance Education. Proceedings
of the IEEE FIE Conference, November.
Nevertheless, this study only considers NN with
backpropagation algorithm, the potential of other
models and algorithms are not discovered. In addition,
due to the nature of NN’s implementation regarded as
“black box”, i-Neuro can be integrated with other AI
techniques such as Expert System, Case Based
Reasoning and Fuzzy Logic that provide reasoning or
explanation facility for each decision made.
REFERENCES
Barker K., Trafalis, T. & Rhoads, T. R. (2004).
Learning From Student Data, Proceedings of the
2004 Systems and Information Engineering
Design Symposium. Jones, M. H., Patek, S. D.
and Barbara Tawney, E., eds., pp. 79 – 86.
Carlson, S. (2000). Neural Networks may transform
college planning. Washington: The Chronicle of
Higher Education, Vol. 46, Issue 29, issue dated
March 24.
Gonzalez, J. M. B. & Des Jardins, S. L. (2002).
Atificial Neural Network: A New Apparoach to
Predicting Application Behaviour. Research in
Higher Education, Vol. 43, No. 2. April., pp. 235
-258.
Hall, O., P. (2002) Artificial Intelligence Techniques
Enhance Business Forecasts: Computer-based
analysis
increases
accuracy
[online].
http://gbr.peperdine.edu/022/print_intelligence.ht
ml
Hammond, D. L. (2000). Teaching quality and student
achievement: A review of state policy evidence.
Education Policy Analysis. 8(1).
Hanke, J., E. & Reitsch, A., G. (1998). Business
Forecasting (6th Ed). New Jersey: Prentice Hall.
Hanushek, E. (2001). Economics of Education. Paul B.
Baltes & Neil J. Smelser (Eds.). International
Encyclopedia of the Social and Behavioural
Sciences, vol 6. Amsterdam: Elsevier Science,
pp. 4200-4208.
Hardgrave, B. L. C.; Wilson, R. L. & Walstrom, K. A.
(1994). Predicting Graduate Student Success: A
Comparison of Neural Networks and Traditional
Techniques. Computers and Operational
Research. 21(3). pp. 249-263.
Jingtao Yao & Chew L. T., (2001) Guidelines for
Financial Forecasting with Neural Networks,
Proceedings of International Conference on
Neural Information Processing, Shanghai, China,
ms757-761
JPA. (2005). <http://www.jpa.gov.my >
Kaastra, I. & Boyd, M. (1996). Designing a Neural
Network for Forecasting Financial and
Economic Time Series. Neurocomputing, 10:
215-236. Elsevier Science B.V.
426
Intelligence Applications in Industry. Kuala Lumpur,
Malaysia. 24 – 25 June. #63. pp. 1 – 6
Siraj, F. & Yusoff, N. (2004). Information System for
Lembaga Biasiswa Negeri Kedah, Universiti Utara
Malaysia, Kedah.
Siraj, F. & Yusoff, N. (2005). Model Peramalan
Pinjaman Pendidikan Negeri Kedah Menggunakan
Pendekatan Rangkaian Neural. Universiti Utara
Malaysia, Kedah.
427