CCE Theses and Dissertations

Date of Award


Document Type


Degree Name

Doctor of Philosophy in Information Systems (DISS)


Graduate School of Computer and Information Sciences


Maxine S. Cohen

Committee Member

Sumitra Mukherjee

Committee Member

Steven D. Zink


Credit scoring is a mathematical means of summarizing a consumer's credit and financial history into a three-digit number. This number provides an easy means of identifying and sorting consumer behavior into categories based on their financial history. To select applicants for loans and to set interest rates on loans, banks and financial institutions routinely use credit scoring. Auto insurance companies also use scoring to decide which consumers will be offered auto insurance and to set the price for auto insurance. Despite success in these two industries, scoring does not appear to be effective in the apartment rental industry in picking desirable applicants for apartment rental.

The first phase of this research analyzed the results of using six commercially available credit scores applied in one apartment complex to the task of selecting applicants. This part of the analysis answered the research question: How effective are commercially available credit scores in predicting applicant financial behavior when renting an apartment? This research determined that these six scores are not predictive and possible explanations are given.

Phase two of this research used neural networks to develop a new model using both credit data and other lifestyle data about the applicant. The hypothesis was that the addition of this lifestyle data would improve accuracy in selecting apartment rental applicants over currently available models based only on credit data. This part of the analysis answered the research question: How is the prediction accuracy of a new neural network based credit scoring model improved by adding lifestyle data to the credit report data? This research indicates that accuracy is greatly improved. Three variables were found to be most predictive for the apartment rental decision and these were a) percentage of satisfactory accounts in the applicant's credit file, b) total applicant income, and c) driving record of the applicant. Four areas were suggested for future study and these are a) understanding the underlying human behavior differences that influence apartment financial decisions, b) addition of "fuzzy logic" techniques to the neural network, c) expanding the number of commercial credit models tested and size of the data set and d) effect of geography on model prediction accuracy. This dissertation also examined U.S. information policy and addressed consumer privacy considerations when using non-credit data to select applicants.

  Link to NovaCat