You are on page 1of 8

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

net/publication/224329371

Expert System for Banking Credit Decision

Conference Paper · October 2008


DOI: 10.1109/ICCSIT.2008.31 · Source: IEEE Xplore

CITATIONS READS
11 6,775

3 authors, including:

Mostafa Mahmoud N. Algadi


Islamic University In Madinah Collage of Computer Science & Information Systems Taif University
14 PUBLICATIONS   68 CITATIONS    2 PUBLICATIONS   11 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Mostafa Mahmoud on 04 December 2015.

The user has requested enhancement of the downloaded file.


Expert System for Banking Credit Decision

Mostafa Mahmoud∗ Najla Algadi, Ahmed Ali


Faculty of Computer Science and Engineering Collage of Postgraduate Studies
Taibah University Sudan University of Science and Technology
Madinah Munawwarah, KSA Khartoum, Sudan
Email: mkhaleel@taibahu.edu.sa Email: naglaa2002@hotmail.com

Abstract situation of the over due debts (financial failure) in


Sudan in the last three years is more than 20% and
The problem of credit-risk evaluation is a very increased yearly according the annual report issued
challenging and important financial analysis problem. from Central Bank of Sudan. The reasons for these
Recently, researchers have found that expert systems problems are mainly due to the lack in using good
perform very well for this complex and unstructured credit ratings and assessment, which might indicate a
problem when compared to more traditional statistical high risk of defaulting on a loan, and thus leads to high
approaches. Expert systems with explanation for interest rates. In other words, there is a financial
decision making can achieve a high predictive failure, because of the weakness in credit-risk
accuracy rate; the reasoning behind how they reach assessment by the bank granting the credit. Those
their decisions is not readily available. This paper problems occurred to the decision maker because the
presents an Expert System for Evaluating and great number of factors that should be considered with
Supporting Credit Decisions on the Banking sector different weight according to each case and also the
(ESESCDB) uses the credit rating weights for each lake of existence knowledge. This financial failure
factor that affecting the decision of the credit. This mainly is due to the lack of experts in banking domain.
work has established an expert system tool that aids To solve this problem an expert system has been
the decision maker to issue the right decision with suggested to keep the related knowledge to be used as
familiar and easy-to-use interface. There are two main a decision support system. The primary goal of the
methods have been applied to acquire the knowledge of expert systems is to make expertise available to
credit evaluations systems in banking with decision makers and technicians. Expert knowledge is
effectiveness, efficiency and correctness, they are a combination of a theoretical understanding of the
direct and indirect methods. The knowledge has been problem and a collection of heuristic problem-solving
verified and evaluated with other senior experts, and rules that experience has shown to be effective in the
then some modifications and enhancements have been domain. Many Knowledge Elicitation methods have
done to reach the final system. been used to obtain the information required to build
the system. Document analysis and decision analysis
1. Introduction are used to classify entities within a domain.
In the following sections, the theoretical base of
The banks face very challenging and important credit decision described in section 2, then the previous
financial problems, because there is a weakness in work in expert systems in financial planning and credit
credit-risk assessment and completion of loan package. presented in section 3. The expert system for banking
The banks crises studies in different countries, whether credit decision is described in section 4. In section 5
advanced or developed, point out that most of the the evaluation of the tool is presented, and finally the
countries which are exposed to financial crises are due conclusion is in section 6.
to the main reason of the over due credits (financial
defaults). It is worth mentioning that 131 countries 2. Theoretical base of credit decision
suffered from bank crises, which differ in its
magnitude during the period (1975-2000). The


On leave of absence from CLAES, ARC, Cairo, Egypt.
There are different kinds of risks that are facing accounting. One trend in the latest work is to merge
banks that may be divided into four kinds according to both knowledge-based and evolutionary approaches.
the classical division: financial risks, operational risks, He assures that opportunities exist to reuse the
business risks, and event risks. In spite of the knowledge bases built for the financial accounting
importance of all these types of risks that must be work for speculative investing. There are many
taken care of by managements in banks, yet research and application in the financial planning and
concentrate has always been on credit risks as it credit field that used expert system as a tool to solve
represents one of the main risks like market risks, many problems in this field.
operational risks, and liquidity risks. Assessing risks is Artificial neural networks have been introduced to
an ongoing process. It typically involves three steps: develop objective expert systems in credit risk [5]. A
First, a bank may engage in a rigorous analytical neural network is ‘‘trained’’ using historical repayment
process to identify risks and, where possible, to experience and default data. Structural matches are
quantify them. In the event risks cannot be quantified, found that coincide with defaulting firms and then used
management may still identify how potential risks can to determine a weighting scheme to forecast
arise and the steps it has taken to deal with and limit probability of default. Each time the neural network
those risks [1,2,3,4,5,13]. Bank management should evaluates the credit risk of a new loan opportunity, it
form a reasonable and defensible judgments of the updates its weighting scheme so that it continually
magnitude of any risk with respect to both the impact it ‘‘learns’’ from experience. Thus, neural networks are
could have on the bank (including the maximum flexible, adaptable systems that can incorporate
potential impact), and the probability that such an changing conditions into the decision-making process.
event will occur. A second step in assessing risk is for However, the network may be over-fit to a particular
the board of directors or senior management to database if excessive training has taken place, thereby
determine the bank’s risk tolerance, based on an resulting in poor out-of-sample estimates. Moreover,
assessment of the losses the bank can afford to sustain neural networks are costly to implement and maintain.
in the event a given problem materializes. Finally, Because of the large number of possible connections,
management can compare its risk tolerance with its the neural network can grow prohibitively large rather
assessment of the magnitude of a risk to ascertain if the quickly. Neural networks suffer from a lack of
risk exposure fits within the tolerance limits. transparency. Since there is no clear economic
interpretation that can be attached to the hidden
3. ES Approach in financial planning and intermediate steps, the system cannot be checked for
Credit plausibility and accuracy. Structural errors will not be
detected until probability of default estimates become
Expert Systems (ES), which are sometimes referred noticeably inaccurate.
to as knowledge based decision support system, are an
application of AI that helps decision makers make 4. ES for Banking Credit Decision
better decisions. ES approach has been suggested to fill
gabs that appeared when traditional approach was used There are two main activities to develop the Expert
in many applications. Such subjects as the general System for Evaluating and Supporting Credit
classifying of ES [7], the various structural Decisions on the Banking sector Tool (ESESCDB).
characteristics of ES, the install phases and difficulties The first activity is to acquire the knowledge to build
of ES, the implementation conditions of ES, and the the knowledge base in the development environment.
implementing successfully in field of ES. There is an The second activity is to integrate the knowledge base
innovative, experimental approach to a literature with the user interface in the consultation environment
review by Rada R. [6], it begins with queries for (Figure 1). In the development environment the
finance-related articles to the ESs with applications knowledge engineer analyze the available document
literature database. He found a classification language including periodicals, references and books, ranks
that is constructed, and retrieved articles are systematic reports and publications, central bank publications and
indexed. He found also that hypotheses are offered policies, researches, working papers, and banks
about the patterns in the distribution of indexed studies; then meet with the domain experts through the
concepts. Results include that authors tended to use knowledge acquisition sessions to verify and elicit the
expert systems tools in the early 1990s but knowledge related to the system from the expert. This
evolutionary computation tools in mid-2000s. stage is a bottleneck part of building such system.
However, the most common financial application area Consequently the knowledge engineer used the
in both the earlier and later periods was financial appropriate knowledge acquisition techniques while
Consultation Environment Development Environment

Knowledge

-Researches, working papers, and banks studies.


Mathematical
Inference base
Models

-Central Bank Publications and Policies.


System Document

-Banks reports and publications.


Analysis

-References and books.


Dynamic Classified

-Periodicals.
User Interface

Database Knowledge Domain


Acquisition Concepts

Explanations Explanation
Modules Knowledge Domain Knowledge
Base Experts Engineer

Figure 1 Expert System for Banking Credit Decisions

meeting with the domain experts to elicit and evaluate most fundamental constituent of a knowledge base. By
the knowledge including the explanation part. During domain ontology, we mean a full declaration of
those meetings the design was establish. concepts, relations between concepts, and relations
between properties. This part of domain ontology
4.1. ES Development Environment represents the lower layer of domain knowledge.
Domain models embodies the upper layer, and
In developing process, we used CommonKADS represent knowledge of what are the domain relations
methodology [2]. CommonKADS follows a systematic required to fulfill a specific function within the domain
approach in knowledge engineering that adopts the of application.
concept of modeling in the process of knowledge base To acquire the domain ontology, we prepared two
system development. CommonKADS methodology editors: Concepts editor, and Relations editor. The
was chosen due to the fact that most current knowledge concepts editor is used to declare domain concepts,
representation approaches are inadequate for the along with their properties, and properties facets. The
systematic development of large knowledge base relations editor, uses the concepts declarations as input,
systems, as they do not strictly separate the analysis and facilitates defining relations between concepts, and
from the design and implementation phases. This relations between properties. Currently we have five
methodology widely used in CLAES+ [10] and distinct representations of relations: Rules, Tables,
achieves a great success in developing many systems. Mathematical functions, Constraints, and Object
The proposed development environment consists of hierarchies. Each of these representations has its own
document analysis and knowledge acquisition internal structure that suits different kinds of domain
processes, that facilitate building and maintaining relations. The selection of the appropriate
knowledge components libraries and mathematical representation of a specific domain relation is up to
models libraries, and constructing knowledge based knowledge engineer, according to some criteria that
systems and mathematical model out of these libraries. characterizes each representation. It is noteworthy that
To build the infrastructure for a large scale knowledge the proposed architecture is not closed to the
based system, we start with acquiring and classifying aforementioned knowledge representations; the
the domain ontology from different sources that is the knowledge engineer is free to add new knowledge
representations, and inference mechanisms that work
+ on them.
Center Laboratory for Agriculture Expert Systems, ARC,
MOLAR. Egypt.
Fiscal Policies, credit policies, rate of exchange policy,
4.2. Knowledge Acquisition public finance policy and government pricing policy.
(4) The Financial Analysis, which examine the
There are two main methods have been applied to financials of the company, financial analysis for short
acquire the knowledge of credit evaluation systems in term loans, financial analysis for long and medium
banking with effectiveness, efficiency and correctness, term loans. The short term analyzes client current
they are direct and indirect methods. Indirect methods centre, the general current centre, the activity cycle and
are used in order to obtain information that can not be sources of working capital structure. The long and
easily expressed directly. Document analysis and medium term analyzes the financial structure, the
decision analysis are used to classify entities within a earning power, the market trends and contingent and
domain. These methods attempt to determine how the expected obligations.
expert makes their decisions. (5) Credit Study Analysis, which has the personal
Document analysis involves gathering information interviews and the field visits and inquiring about the
from existing documentation. This process has been client performance with the bank and other Banks.
achieved through interaction with a human expert to (6) The Collaterals: The last one is the collaterals, it
confirm or add to this information. The knowledge were classified to eight kinds the Bank guarantee, real
engineer who is one of the authors of this research help Estate, deposit account, promissory notes, holding
in eliciting the knowledge from deferent sources such company guarantee, storage of the product, personal
as: Periodicals, References and books, Banks reports guarantee and over Draft Guarantee.
and publications, Central Bank Publications and
Policies, and Researches, working papers, and banks 4.2.2 Knowledge Elicitation from the Experts.
studies. That knowledge contains mathematical and Analyzing and filtering the previous criteria's with the
statistical treatment of data for correlation analysis, experts, then we reached the following five
trend analysis, hypothesis testing, and predictive conclusions criteria taken on to the consideration in the
analysis. Direct methods or interaction with the domain analysis that covering all the decision-making aspects
expert involve directly questioning a domain expert on either qualitative or quantitative and it has obvious sign
how they do their job; the domain expert has to be in credit process. It contains the first and second stages
reasonably articulate and willing to share information. of analysis criteria's with new arrangements suitable
The knowledge has been verified with other senior with the last classification: the knowledge elicitation
experts, and then some modification and enhancement from expert provide new credit ratings: Economic
have been done to reach the verified system. Feasibility Study, Financial Feasibility Study,
Technical Feasibility Study, Marketing Feasibility
4.2.1 Document analysis. According to Basel [1], Study, and Collaterals.
Hashad [13], and others [2,3,4,5] credit ratings assess
the credit worthiness of an individual, corporation, or 4.2.3 Validating and Verifying the Knowledge from
even a country. Credit ratings are calculated from the domain Experts. The extracted and elicited
financial history and current assets and liabilities. From knowledge have been validated, verified, and updated
final accounts, market trends, published statistical with the domain experts and they also provided other
reports. Six criteria taken on to the consideration: the knowledge's to complete it, the experts' contribution
feasibility study, the client study analysis, the general appeared in different aspects of the system.
circumstances analysis, the financial analysis, the
credit study analysis, and the collaterals. 4.3. Knowledge Base Components
(1) The Feasibility Study, which consists of economic,
financial, technical, marketing, social and environment The knowledge base for Evaluating and Supporting
feasibility. In addition, every one from them has many Credit Decisions on the Banking sector Tool
pointers and questions. (ESESCDB) consists of five main components, namely
(2) The Client Study Analysis, which contains the Economic Feasibility Study, Financial Feasibility
client five Cs Merit Creditworthiness (characters, Study, Technical Feasibility Study, Marketing
capacity, circumstances, capital, and collateral). Feasibility Study, and Collaterals as depicted in figure
(3) The General Circumstances Analysis, which has 2. In the credit, ratings, which are used in the system,
two lines international and national. The global one hundred points are suggested as a total weight, and
contains inflation and stagnancy, the quality and then every part of the five components has default
quantity protection policy and economics amount of points, which have been discussed with the
agglomerates. The national situation has Financial and experts. The economic feasibility study has fifteen
points. Each one of these ratings and its contents are flow, financial analysis, FIRR, NPV, sensitivity
divided its points in the suitable distribution. Financial analysis, break-even analysis, and profit trends with
Feasibility Study has thirty points. Technical time series. The qualitative evaluation consists of client
Feasibility Study has twenty points. Marketing assessment, customer relation ship with the bank and
Feasibility Study has twenty points. Collaterals have other banks, and other special factors affecting the
fifteen. In those five components there are some project.
critical factors that may affect the decision making Technical Feasibility Study is categorized in six
process in all the parts of the system, the project will major groups – the project specifications, costing,
be rejecting as soon as it happened. There are many production capacity, the technical expenditure &
important factors those affecting the decision to give training, availability of after sale service, production
recommendations to grant the credit, or reduce it, or management, and environment.
recommends reviewing the credit application before Marketing Feasibility Study is categorized in four
granting, and lastly rejects the credit application if it major groups: supply & demand and gap analysis,
did not meet the criteria of evaluation, and there are pricing, quality of the product, swot analysis strength
less important factors in the system, in this way we & weakness (internal environment of the company),
used different approaches to reach the same result. The opportunities & threats (external environment of the
following subsections describe each component. company), and marketing policy. Each group is
evaluated a number of factors.
Evaluation of Collaterals is categorized in eight major groups-
Financial
feasibility bank guarantee, deposit account, real estate,
Evaluation study Evaluation promissory notes, holding company guarantee, storage
of Economic of Technical
feasibility feasibility of the product, personal guarantee, over draft
study study guarantee.
There are some critical factors that may affecting
Credit
the decision making process in all the parts of the
Ratings and system, the project will be rejecting as soon as it
Viability
happened. On the other hand there are many important
Evaluation of
that affecting the system. By using those factors we
Evaluation of
marketing
the collaterals
want to sustain that the project has a sensitivity
feasibility
study analysis and it is conducted under three scenarios.
Earnings power, it is very important factor as it
represent the first defense line for the project.
Figure 2: Knowledge Base Components
Assessment of the client, in this assesses explaining the
The economic feasibility study is categorized into client worthiness which is very important factor in
three major groups: global economic environment, credit evaluation process. The less important factors in
national economic and political situation and social the system have been identified to allow the system to
impact. Each group is evaluated through examining the use different approaches to reach the same results.
related factors. For example in the social impact, there
are towlines of evaluation the first are the sector 4.4 Mathematical Model Components
situation (micro) and the social impact (economic
internal rate of return). In the sector situation (micro) A library of mathematical models has been
the evaluation is examined through the laws of the integrated with the system to select the appropriate
country if it allows for the establishment of the project model during the run time according to each situation.
and the investment climate suitable for the project. The The library contains all modules that necessary to
social impact (economic internal rate of return), here support the expert system with the result value of each
the assessment consists of the EIRR ratio in module. Some of those mathematical models have
comparison with the cost of capital. This part is different methods to achieve the goal according to each
achieved throughout forty questions; each one has situation and available data, and other using simple
different weight according to the degree of it's effect of way. For instance to evaluate the credit there are many
the evaluation. factors should be considered such as cash flows, that
Financial Feasibility Study has two major groups should be evaluated by considering all of the cash
have identified, quantitative evaluation and qualitative inflows and out flows induced by investment decision
evaluation. Each one of these groups has special in question.
contents. The quantitative evaluation contents are: cash
4.5 Explanation Knowledge Base Components accomplish this task the how-solution module starts
with the user-selected conclusion and collects all rules
At the end of consultation session the end user can that contain this particular conclusion in the current
ask “How” the system reaches the generated solution. inference step and the related derived values in all
Expert system shells provide this type of explanation previous inferences steps. The second step is to
as tracing the line of reasoning; the end user cannot compare the premises in each rule with the finding in
understand this explanation. Most system justify to the the system, to get the rules that their premises have
user how it reached this solution by displaying the matched with the system findings. This step is very
suitable text that links the findings with the conclusion. necessary to reject other rules that are not related to
If more than one solution path is generated from the this particular conclusion. We consider those matched
given finding the system will display all those paths rules and get the explanation attached with them, and
with their explanation text. Consequently there is a store the derived values in the premises in each rule (if
need to generate a special knowledge representation any) to be used in the previous inference step at the
schema and problem solving method to generate the next loop execution to get the rules that are related to
explanation that acquired from the domain expert. the user-selected conclusion until reaching the first
There are two solution approaches that can satisfy inference step.
the above specifications. The first is to keep the history
of the session in a temporary structure in the memory 5. Evaluation
or database. This session history used to know how the
path followed through to reach the specified In order to demonstrate the applicability of the
conclusion, and then construct the suitable explanation presented architecture, although the system has been
during the consultation session. The second alternative tested using a lot of hypothetical cases provided from
is to use the meta knowledge to construct the the domain experts, the expert system was applied in
explanation when the user asks “how-solution” of a two real cases to measure the impact of using the
conclusion while there is no overhead in running system to evaluating and supporting credit decision on
procedures during the consultation time to construct the banking sector. The methodology followed to
this explanation. achieve this objective was applied: 1) Determine two
The second alternative uses the meta knowledge different banks that have real cases they apply a study
that is constructed only one time as described in the in addition to the decision in each bank. 2) Collect the
previous section. But in the first alternative we keep information needed to run the proposed system for
the session history during the consultation time, even if each case (some information may not be available).
the user doesn’t want any explanation at all. The Note that in this step the needed information to run the
second alternative is more suitable to construct the system is collected from the bank, but not the result of
needed explanation because the first one has more the study arrived at by each bank. This is to make sure
overhead to construct the explanation while running that, there is no bias in advance or effect of the existing
the expert system. In addition, we can’t generate this result on the tests that will be made by the proposed
kind of explanation from the expert system because it system. 3) Run the system for each case by providing
needs deep knowledge to justify the conclusion result the information to the system, in order to get the result
that does not exist in the knowledge base. So when for each case and document in the form of a report. 4)
building the ‘How Solution’ explanation we need to Show the study result that applied in each bank. 5)
know all conclusions results or derived values in each Compare the result for each case with the result of the
inference step. The derived values will be used in the system. In first case the decision which was taken by
intermediate result in the inference steps. But the final the bank compatible with the proposed expert system.
result (the final conclusion) exists only in the last Although the system agreed with traditional system
inference step. The meta knowledge that already but give justifications for it is recommendation consist
captured from the expert during the knowledge of the negative points that must be taken into
acquisition process for each conclusion pass such as consideration. The system in the second case gives the
inference name, inference order, inference role, decision that is compatible with the bank decision both
relation, rule fired, generated text explanation, and of them give decision granting credit. The system
expert explanation will be used to construct the support the bank decision by calculated different
explanation. equations in financial analysis helps in measuring the
There is a module of “How Solution” explanation to quantitative data then helps in assessing the credit
determine the exact explanation for each specific case. decision process. The system provides the explanation
The result of this module is a justification of why the to support the conclusion that reached. One important
system reached that particular conclusion. To
issue of using the system that it reduce the time for [3] Oh, K. J., & Kim, T. Y., "Financial market monitoring
evaluating the credit because it provide to the user with by case-based reasoning". Expert Systems with Applications,
the required information for each case, then the user 2007, 32(3), 789–800.
can focus only for those required information to flow [4] Kumra, R., Stein, R. M., & Assersohn, I., "Assessing
logical system which will reduce the risk of over due knowledge based approach to commercial loan
debts rather than gathering a lot of information about underwriting". Expert Systems with Applications, 2006,
the credit under investigation without affecting the real 30(3), 507–518.
decision, also the system provide a conclusion even if [5] Linda A, Gayle D, Anthony S., "Issues in the credit
some data missed. risk modeling of retail markets" in Journal of Banking &
Finance, Volume 28, Issue 4, April 2004, Pages 727-752
6. Conclusion [6] Rada, R., "Expert systems and evolutionary computing
for financial investing: A review." Expert Systems with
The major contribution of this research is to propose Applications, 34 (2008).
an expert system for evaluating and supporting credit [7] Luke Hodgkin son, Ellen Walker, "An Expert System
decisions on the banking sector (ESESCDB). This ES For Credit Evaluation And Explanation" Hiram College,
is designed not only to justify the chosen decision, but POB 67, Hiram, OH 44 - 2003
also to help the bank consultants to feedback their
[8] Mahmoud M., El-Araby K., Rafea, A., "LIMEX: An
experience and to make recommendations to the
Integrated Expert System for Lime Crop Management." 2nd
system. In addition, it allows the user to change the IFAC/IFIP/EnrAgEng workshop on Artificial Intelligence in
values of the system parameters. In this way, it Agriculture, 29 May - 1 June, 1995, The Netherlands.
provides an efficient means to run the consultation
using a different set of inputs. The system can help the [9] Wagner W.P. and Otto J., Chung Q.B., "Knowledge
acquisition for expert systems in accounting and financial
human experts in explaining why a particular decision
problem domains" Department of Decision and Information
is made. It can evaluate the credit-worthiness of Technologies, Villanova University, Villanova, PA 19085
applicant companies or individuals, with recommend USA February 2002.
for a credit line be accepted, rejected, reduced or
revised. The system aids a credit officer in the risk [10] Rafea, A., Edree, S., El-Azhari, S., Mahmoud M., "A
Development Methodology for Agricultural Expert Systems
assessment and completion of a loan package. It can
Based on KADS", Proceedings of the Second World
help in training new employees, to improve Congress on Expert Systems, January 1994.
productivity, to conduct marketing research, and any
number of other functions. However, they also offer [11] NASA, CLIPS v. 6.2 Lyndon B. Johnson Space
considerable advantages to strengthening the analytical Center, Information System Directorate, Software
and Communication skills plus improving research Technology Branch, Houston, Tex., 2002
skills of university students who attempt to design such [12] Schreiber, G.; Wielinga, B. J.; Akkermans, H.; de
systems. Velde, W. V.; and Anjewierden, A., "CML: The
CommonKADS conceptual modeling language." In
Proceedings of 8th European Knowledge Acquisition
Acknowledge Workshop (EKAW), 1–25. 1994.
The authors would like to acknowledge the domain [13] Hashad, Nabil "Your Gide to Banking Credit Resk"
experts team from central bank of Sudan, development BASEL part 2, 1997.
export bank, and Sudanese Banks Association. Thanks
are also due to the expert Dr. Bashier Omer Faddllalh
economist advisor of the Islamic development bank in
Jeddah.
7. References
[1] Basel Committee on Banking Supervision, "Credit
Ratings and Complementary Sources of Credit Quality
Information." No 3, August, 2000.
[2] Nikola A Tarashev, "An empirical evaluation of
structural credit risk models" Bank for International
Settlements- Basel, Switzerland -July 2005. BIS Working
Papers -No 179

View publication stats

You might also like