Award Date

1-1-2005

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Mathematical Sciences

First Committee Member

Rohan Dalpatadu

Number of Pages

48

Abstract

Statistical models are most predictive and robust when the historical data used to build the model covers sufficient representation from the population. A financial organization uses a scoring model, developed with historical data, to determine if a consumer should be granted a line of credit for which they apply. When the existing model is too old or the applicant population has changed, a new model is to be developed, but the range of historical data, essential to the predictive model development, does not include the entire applicant population because of the declines. A technique called reject inference is typically used to infer the unknown performance of the reject population as a new scoring model is developed. This paper explores several techniques that have been published, selects the most promising technique, and introduces an adjustment to the existing technique.

Keywords

Data; Inference; Methodology; Missing; Problems; Reject; Skewed

Controlled Subject

Mathematics

File Format

pdf

File Size

1013.76 KB

Degree Grantor

University of Nevada, Las Vegas

Language

English

Permissions

If you are the rightful copyright holder of this dissertation or thesis and wish to have the full text removed from Digital Scholarship@UNLV, please submit a request to digitalscholarship@unlv.edu and include clear identification of the work, preferably with URL.

Identifier

https://doi.org/10.25669/k1tm-s4a5


Share

COinS