Hybrid Intelligent System for Credit Approval in Banks | Free Essay Example

Hybrid Intelligent System for Credit Approval in Banks

Words: 3136
Topic: Tech & Engineering

Executive Summary

The use of credit by consumers necessitates the use of an approval process by banks to vet suitable candidates to avert lending risks. The major risks a bank deals with when forwarding credit facilities is fraud and default. Banks use credit rating as the first line of defence against credit and fraud associated risks. The risks arise out of information accuracy and procedural challenges.

The use of decision support systems reduces the risks associated with lending. They rely on various disciplines to predict human behaviour and tendencies. Making them web-based expands the scope of the database from a location unit to a global pool. Different Artificial Intelligence (AI) techniques provide the tools required to improve the credit approval process. They include Bayesian Networks, Case-Based Reasoning, Rule-Based Reasoning, and Artificial Neural Networks.

Each of them provides different modes of data acquisition and processing making their input unique to the credit approval process. Bayesian systems provide the best means of background screening to check whether a data set conforms to a certain pattern. Case-Based Reasoning and Rule-Based Reasoning have ‘learning abilities making them useful for trend analysis. Artificial Neural Networks can compare large data sets making them useful for massive fraud detection.

The use of hybrid intelligent systems makes available the best opportunity for alleviating lending risks faced by banks.


Many consumers take advantage of credit facilities offered by banks to meet their objectives. To access credit, each prospective borrower goes through a credit approval process. “Credit approval is the process a business or an individual undergoes to become eligible for a loan or to pay for goods and services over an extended period of time” (Leotta, 2011). Banks use various techniques to screen applicants before they allow them to utilize their credit facilities. The efficacy of the tools and techniques used to screen applicants constitute a significant risk to the banks.

Businesses and individuals that are not creditworthy may receive the nod to utilize credit. This is the source of major problems for banks leading to bad debt, default on payments, and high cost of litigation. Information technology has the potential of enhancing the effectiveness of the credit approval process to reduce the risks associated with the provision of credit. As Hill Associates (2002) state, “telecommunications technology perhaps more than any other technology continually shapes the very fabric of our global society”.

Problem Identification and Analysis

Banks are major providers of credit. This exposes them to various risks associated with lending. These risks culminate in default on loan repayments and fraud-related losses. To mitigate the risks, banks use certain procedures for credit approval, meant to detect fraudulent applications and to foresee the possibility of default. The procedures vary from institution to institution and from product to product, because of the diverse nature of borrowers and the products they require. Normally, the procedures revolve around the applicants’ background and their credit history.

Banks rely on credit scores provided by credit rating agencies as a basic form of screening. Evans (2011) feels that a person’s basic credit score tell them very little and is, therefore, “useless”. The procedures reveal factual inconsistencies in the application form when compared with data from other sources such as the credit reference bureaus. They also use credit history and compare an applicant’s creditworthiness against their current ability to repay. In light of developments in the information technology sector, banks have the potential of improving their credit approval process by using web-based decision support systems.

There are two areas where there are problems in the credit rating process. They are the quality and accuracy of the information availed and procedural challenges. Information related problems arise from the multi-agency nature of data collection for the determination of creditworthiness, and erroneous information provided fraudulently or involuntarily. The information banks use in the credit approval process comes from many sources. They include; the applicant, credit rating agencies, pools of shared information from other banks and lenders, and internally generated sources.

“Banks increasingly use data mining to analyze great quantities of information, to get a feel of trends on potential new customers and fraud patterns (Leotta, 2011). This shows that there is a strong drive towards computerization of the credit approval process. Credit approval has its basis on trust, which a computer cannot measure. Computerised systems rely on ratios and other documented patterns such as default rate on past loans to establishing a credit history. In the UK, banks rely on several sources to determine client credit scores.

They include the application form a client fills, personal history with the bank and files obtained from at least one of the three credit rating agencies. These agencies compile data obtained from the electoral roll containing details of residence, records from courts, records other lenders have based on past applications to them, fraud reports, and data from accounts held by other banks, utility providers, and other organizations. The interagency nature of the information used makes it easier to arrive at a consistent view of a creditor. However, it presents reliability problems whenever any of the agencies involved have erroneous information in its database.

Leotta (2011) states, “information provided by credit bureaus is often incorrect. In a National Association of Independent Credit Reporting Agencies survey conducted in 1994, nineteen per cent of credit reports were found to contain outdated information, and forty-four per cent were found to lack information regarding current balances and payments on existing loans” (Leotta, 2011).

The second major challenge is fraud and dormant information. This challenge in the credit approval process arises from the feeding of wrong information into the system. Fraud misleads banks into providing credit facilities for undeserving entities. While fraud is voluntary, dormant contracts such as mobile phone contracts that are no longer in use, but still listed as active, cause involuntary misinformation to credit officers. They present the borrower as more credit laden than they are. Credit agreements no longer in use contribute to a borrower’s credit history, so long as they are active. This is why Evans (2011) advises lenders to “close down any credit agreements no longer in use”.

At the procedural stage, credit officers use the information provided to calculate the credit score of the applicant. If the score is satisfactory, they forward the credit facility. At first, it may seem like a straightforward process. It is more complicated than that because there is no standardized format for determining credit ratings. Lewis (2010) states, “Scoring systems are never published, and differ from lender-to-lender and product-to-product.

The Oesterreichische Nationalbank (2004) identifies substantive errors and procedural errors as the two types of errors incumbent on the credit approval process. Substantive errors occur when there is an erroneous credit assessment despite full disclosure. Procedural errors, on the other hand, come about because of a lapse in the procedure, usually by the credit officers. Reasons for this type of error include incompetence, incorrect application of the credit approval process and negligent or intentional misconduct by officers.

One additional problem with credit approvals is what Lewis (2010) calls “the rejection spiral”. Here, an applicant gets a series of rejections on credit applications, and after a few of them, the rest of the institutions simply reject the application based on the applicants’ rejection history. The basis for the initial rejection may be valid. It could be an unresolved error. This means that banks lose on good business while lenders who need the credit and maybe qualified loose on the opportunity to get credit. This does not benefit any party.

The credit approval process is a decision making process. Therefore, problems associated with the credit approval process are decision-making problems. The basic problem of the credit approval process is gathering accurate information, and processing that information as effectively as possible to determine the credit risk associated with each application. The solution to this problem lies in decision support systems.

The problem of the collection of accurate information spans the agencies involved in collecting it in the first place, and then all the people and processes involved in processing the information. A solution to it will aim at more accurate data collection and more accurate data processing procedures. Stated more succinctly, “The quality of credit approval depends on two factors, i.e. a transparent comprehensive presentation of the risks when granting loan on the one hand, and an adequate assessment of these risks on the other” (Oesterreichische Nationalbank, 2004)].

Theoretical and Conceptual Framework for Web-Based Decision Support

Decision support systems are “interactive, conversational computer systems supporting decision makers” (Jarvis, 1976). A web-based decision support system utilizes internet capabilities to provide a platform for decision making from any location on earth provided there is someone who can provide the information required for its operation and issue instructions necessary for the system to perform its functions. This is because decision support systems “rely heavily on human intuition, judgment, and experience as an integral part of the decision process (Jarvis, 1976).

All decision support systems rely on management science, computer science, and behavioural science as fundamental knowledge areas for the development of solution frameworks. What a decision support system does is that it “retrieves information from a large data warehouse, analyzes it by user specifications, then publishes the results in a format that users can readily understand and use” (Zopounidis & Doumpos, 2011). Using the internet as the platform promises worldwide access to different pools of information.

This is exactly what banks need to improve their decision-making process when conducting credit evaluations. Currently, banks, because of relying on credit rating agencies for background checks, do not have any information relating to parking tickets, criminal record, child support information, previously declined applications, defaults and missed payments that are more than six years old. This extra pool of information is accessible from internet sources and may serve to improve the quality of information that banks use in the credit approval process. After all, decision support systems rely on the accuracy of information to provide useful results.

The form a typical decision support system take includes, “the database, the model base, and the user interface” (Zopounidis & Doumpos, 2011). For a web-based decision support system, the database in use exceeds a single computer or even a Local Area Network (LAN). It is a worldwide database accessed through many servers in diverse locations. The model base includes the methods and techniques applied to execute the functions of the decision support system. It is responsible for, “performing all tasks that are related to model management, such as model development, updates, storage, and retrieval” (Zopounidis & Doumpos, 2011). The user interface completes the connection between the database and the model base, by giving the user a platform from which to take advantage of the system’s capabilities.

“The most common AI tools used for decision support are Bayesian Networks, Artificial Neural Networks, Case-Based Reasoning, and Rule-Based Systems” (Burnside & Kahn, 2004). Bayesian Networks consist of a structure, probabilities, and an inference logarithm as the basis for its operations. The structure consists of a set of nodes that represents the relationships between data sets. The determination of the probabilities involves the use of expert opinion and takes time to develop. Once determined, an inference algorithm based on the data set provided forms with the basis of the network.

The developed system can handle fresh data very quickly giving inferences on what the data means, but cannot handle new variables such as emerging trends. This makes Bayesian systems suitable for well-established products with established trends patterns. It is good at inference as an AI system, but cannot “learn”. Artificial Neural Networks (ANN) provides another framework for the development of a web-based decision support system for use in credit approval. They use neural networks and they are very useful for deriving relationships between various data sets. They would be invaluable for fraud detection because they can quickly show the relationships between different applicants and application forms, which can escape the human eye.

They also provide a means of discovering emerging patterns in the credit market. The third artificial intelligence technique available for consideration is Case-Based Reasoning (CBR). Case-Based Reasoning uses stored data to analyze input data by comparing the features of the new input with what the database holds. The database is the source of learning for the system. This technique provides a way of checking whether a new application form has elements of risk or opportunity that the bank has dealt with before. The major attraction in the Case-Based Reasoning system is that it can learn by the addition of new data to the old database.

The final artificial intelligence technique available for use in a web-based decision support system is the Rule-Based System. This system consists of “a knowledge base in the form of production (IF_THEN) rules, an inference engine (an algorithm used to apply those rules), and an explanation method that allows the user to interact with the knowledge base” (Burnside & Kahn, 2004). It uses forward chaining, which starts from facts and works towards an inference, and back chaining, that moves from an inferred position to generate the causative facts.

This system is ideal for determining which initial conditions such as application information will result in a successful credit relationship. Through back chaining, the system can infer the desired starting conditions for the outcome a bank desires. Such information feeds the marketing effort. The rules require expert development based on the experience of people who understand the workings of the credit market.

After the development of the system, the rest of the infrastructure remains standard for internet support. They include servers, ports, routers, IPs, gateways, and pathways.

Relevant Decision Making Support and AI Techniques

We have outlined two general classes of problems. One class is those associated with data collection. They have to do with comprehensiveness and accuracy. The second class is the procedural issues in handling applications, based on the information collected. The solution proposed encompasses a large data acquisition and processing capacity on one hand, and an enhanced information processing system using AI decision support systems for the credit officers.

The internet houses loads of data on individuals. It is possible to gather information on a potential client, using search engines. Using similar algorithms, banks, either alone or in tandem with others have the opportunity of developing their search engines to provide them with information on all aspects of a clients application. What they miss from credit rating agencies, they can acquire on their own. The essential characteristics include an algorithmic arrangement that filters information retaining what the banks need to develop an opinion on someone’s credit habits. There are certain problems though such as people with multiple names, fake online identities, or simply, online absence. Essential background information such as crime data is easier to find because third parties generate them. While not entirely reliable, this type of information will provide an additional layer of security checks and will provide a basis for raising red flags on certain persons.

The second element of the proposed solution is the development of a robust decision support system for use by credit officers. No doubt, they already utilize decision support systems but they can do with improvements. The proposed system reduces human error in deciding to forward credit by detecting inconsistencies between a decision made and earlier ones, with the gap showing a possible deviation from the norm. Such a system would rely on Case-Based Reasoning, which does a good job at detecting deviation from logged experiences.

Combined with Artificial Neural Networks, the power to detect fraudulent applications will increase tremendously. While Case-Based Reasoning systems provide a comparison between specific cases against previous ones, Artificial Neural Networks provides comparisons with whole data sets, thereby making it possible to detect massive fraud targeted at the institution. The system’s primary contribution would be to screen new applications against the past application to determine similarities with fraudulent applications discovered in the past.

The third desirable element is the capacity to predict the possibility of a good credit relationship based on preferred initial conditions. Rule-Based Systems have this capacity. They will use the ruleset to compare an applicant’s credentials with those that match ideal credit clients. This makes the banks better equipped to decide whether the business promised is worth the risk. This proposed system promises to reduce human errors in the credit approval process and providing decision-makers with different comparisons to aid their decision.

The fourth element of the system is an online component that allows for counterchecking with other online sources. This will provide banks with much more power than presently possible by only relying on credit rating agencies. Bayesian systems will best fit as frontline defences while conducting background searches on clients. They best capture standard issues and will provide the best means for prequalification as a first stage in determining which applications to process.

A hybrid system incorporating the four elements above will improve the screening process using Artificial intelligence techniques and will reduce the risks associated with lending. As Servigny and Renault (2004) attest, “tools and techniques for managing credit risk have become both increasingly sophisticated and more readily available”. The system will provide banks with greater autonomy when deciding on potential clients and will improve the accuracy of the information they have at their disposal. It allows banks to develop their standards relating to who qualifies to be a lender, not necessarily based on the opinion of the credit rating agencies.

It reduces the errors that a credit officer may make, thereby increasing the confidence in decisions made by them. This risk reduction promises better business for the banks and better service for potential customers including alleviating frustrating eventualities such as the “rejection spiral”. The process will also free banks from the traditional fixation on stability characterized by ownership of fixed telephone lines rather than mobile lines, long-term employment history, staying in one place for long preferably in an owned house and a long-term relationship with a single bank. (Evans, 2011). People are increasingly mobile, yet their presence on the internet makes them static.

Recommendations and Conclusions

For banks to solve their credit rating approval problems, they have the opportunity of utilizing AI tools and techniques.

  1. To solve application form and information quality-related problems, the use of Bayesian systems running on a web-based platform holds the best promise. Information required for decision-making must be, “accurate relevant and complete” (Power, 2002).
  2. The identification of trends for fraud prevention Rule-Based reasoning and Case-Based reasoning provides the best opportunities for flagging risky applicants and risky trends.
  3. Also, Artificial Neural networks provide the opportunity for banks to detect large-scale fraud in the applications they receive.

A hybrid system is the best way to go since some of the techniques provide information relating to large data sets while others explore individual data sets to provide information on risky accounts.

Reference List

Burnside, E.S. & Kahn, C.E., 2004. Artificial Intelligence Helps Provide Decision Support in Radiology. Web.

Evans, T., 2011. Improving Your Credit Rating. Web.

Hill Associates, 2002. Telecommunications: A Beginners Guide. Berkeley: McGraw-Hill/Osborne.

Jarvis, J.J., 1976. Decision Support Systems: Theory. Final Report. Ohio: Online Information for the Defence Community Battelle Columbus Labs.

Leotta, J., 2011. Credit Approval. Web.

Lewis, M., 2010. Credit Rating: How it Works and How to Improve it.. Web.

Oesterreichische Nationalbank, 2004. Guidelines on Credit Risk Management: Credit Approval Process and Credit Risk Management. Vienna: OeNB Printing Office.

Power, D.J., 2002. Decision Support Systems: Concepts and Resources for Managers. Westport CT: Greenwood Publishing Group.

Servigny, A. & Renault, O., 2004. Measuring and Managing Credit Risk. Illustrated ed. New York, NY: McGraw-Hill Professional.

Zopounidis, C. & Doumpos, M., 2011. Decision Support Systems. Web.