Display Full Text | Download Full Text
441.57 kB PDF file

Degree

MS, University of Cincinnati, Engineering : Computer Science, 2001.

Abstract

In this thesis we address the problem of mining association rules from databases containing quantitative attributes. Values of many quantitative attributes are distributed as strong concentrations in a few narrow regions across a very wide complete range of possible values. Intervals of equal width for such domains are not meaningful and may miss out on many peculiarities of data. Our methodology consists of a two phase approach. The first phase determines the boundaries around most meaningful intervals of the value range. We seek to maximize the information content of the choice of the selected initial interval boundaries. The second phase executes the association rule mining algorithm which uses these interval boundaries and modifies them, if needed, to determine rules with specified support and confidence levels. The set of generated rules is then examined to keep only the most specific versions by deleting their more general versions from the set. We have run tests with this algorithm using a network traffic database and the results obtained are presented in the thesis. We also contrast the benefits of this approach with the one in which we may start with uniform width, fixed size, intervals for a quantitative attribute.

Subject Headings

Computer Science

Keywords

data mining; quantization

Advisor

Raj Bhatnagar

Document number: ucin983500840
Permalink: http://rave.ohiolink.edu/etdc/view?acc_num=ucin983500840

★

This ETD has been downloaded 576 times (through March 2013)

This ETD was one of the top fifty most frequent downloads of 2002.