The report contains a concise definition and analysis of the problem and the goals we hope to achieve.

To be able to bring this project to a good end, we will need to set the scope and environment of the problem itself first. We decided on considering a large, public bank (e.g. Mellon Bank or PNC Bank) that will be screening individual applicants who are seeking a personal loan. At this point we do not wish to set the amount range of the loan, this can be determined after further research. The decision-maker will be a loan representative or a committee of representatives of the bank.

The main objectives for the bank will be to make a profit. To gain as much profit as possible the loan applicants have to be screened successfully, which can be seen as a sub-objective. Another sub-objective is to be customer service-oriented, which results in a happy clientele (and more profit!).

In order to address these objectives, we have to define the central issue of our problem. In this case this is: to determine whether or not the loan applicant is eligible for credit at the specified bank. Clearly, a customer that is granted a loan but then defaults will not be of positive influence on the profit of the bank.

Initially, our decision options (i.e. the recommendations to the loan-representative(s)) are going to be simply "yes, the applicant is qualified to receive credit" or "no, the applicant is not qualified". After doing more extensive research, we may expand the options to also have: "yes, the applicant is qualified, but with certain restrictions or stipulations" or "no, the applicant is not qualified: the reasons are as follows… and some recommendations to receive credit in the future are…"

The bank ultimately wants to make a profit, however, if they offer exorbitant interest rates, they will detract from their customer base. Likewise, the bank wants to attract as many customers as they can by offering a low, competitive interest rate, but not so low as to detract from the bank’s profit. Another conflicting objective is the bank’s quest to give as many loans as possible (for the higher goal of profit), but to screen applicants wisely at the same time so as not to choose high-risk customers (i.e. high probability of defaulting on loan payments, therefore diminishing the bank’s profit). Also, the bank wants to collect as much information on the applicants as possible, but has to be careful about sensitive information such as race and age. An applicant could have grounds to sue the bank based on the Civil Rights Act (discrimination against race, sex, age, etc) if the applicant is denied a loan.

The initial model included in this report shows the first thoughts about the credit-worthiness problem. It contains a lot of nodes representing sources of uncertainty. Some of the key uncertainties are:

• the current and potential state of the economy
• possible accident/illness to applicant
• the future income potential of the applicant
• applicant’s possible loss of job
• the applicant’s integrity

Historical data and an expert opinion could be utilized to reduce all of the above sources of uncertainty. Also, the applicant’s health history and all of the risk factors in the applicant’s life could be used to determine the probability of accident or illness. The applicant’s educational background, his/her current job, and the state of the economy could be used to determine future income potential and also potential job loss. The applicant’s credit history and his/her criminal record could be examined to verify the applicant’s integrity.

The dark nodes represent a simple decision problem. One decision node, one chance node and one value node. The value node represents the profit the bank makes by granting a loan or not. Keep in mind that maximizing the profit is the main objective of the bank.

The main goal of our project will be to create a requisite, realistic decision model to determine an applicant’s credit-worthiness. We hope to achieve this goal by doing extensive research in real-life situations and applying the techniques we learn during the course to our model.

The second not unimportant goal is to develop a sample application for our decision model. The application should be able to give decision support to the decision-maker (the bank-representatives in this case), based on the appropriate screening information. It should have an easy to understand user-interface.

- - - - - - - - - -