Skip to Main Content
Frequently Asked Questions
Submit an ETD
Global Search Box
Need Help?
Keyword Search
Participating Institutions
Advanced Search
School Logo
Files
File List
14040.pdf (1.14 MB)
ETD Abstract Container
Abstract Header
Study of Effect of Coverage and Purity on Quality of Learned Rules
Author Info
Gandharva, Kumar
Permalink:
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1428048034
Abstract Details
Year and Degree
2015, MS, University of Cincinnati, Engineering and Applied Science: Computer Science.
Abstract
Rule based algorithms have emerged as a highly effective classification technique with a wide range of applications in the field of medicine, financial data analysis and business marketing to name a few. These classifiers work with real world data and are used to make predictions based on high purity rules developed using pattern mining algorithms. There are a number of aspects that differentiate rule learning algorithms from each other. One way to differentiate them is based on re-use of training instances in rule induction. Existing rule based techniques either do not allow sharing of training instances, discovering too few patterns or allow unlimited sharing of training instances, generating an explosive number of patterns. Recent rule induction algorithms which focus on controlling instance re-use, fail to draw a relation between performance of the classifier and extent of instance re-use. In this work, we propose a novel approach to generate high purity rules by restricting how many times an instance can be utilized while mining frequent patterns. In order to avoid generating an explosive number of rules, we introduce a parameter known as Coverage Limit, to allow control over contribution of each instance in the data towards rule generation. We study the effect of varying the Coverage Limit and Rule Purity in order to achieve best classification accuracy. In addition to this we also propose a Weighted Voting technique which allows multiple rules to collectively predict the label of an unseen instance. A detailed analysis of results on several datasets confirms that the proposed method performs better at classification than many existing techniques.
Committee
Raj Bhatnagar, Ph.D. (Committee Chair)
Yizong Cheng, Ph.D. (Committee Member)
Carla Purdy, Ph.D. (Committee Member)
Pages
72 p.
Subject Headings
Artificial Intelligence
Keywords
Rule coverage limit
;
Control Instance re-use
;
Multiple coverage
;
Control rule purity
;
Rule based classification
;
Classification using multiple rules
Recommended Citations
Refworks
EndNote
RIS
Mendeley
Citations
Gandharva, K. (2015).
Study of Effect of Coverage and Purity on Quality of Learned Rules
[Master's thesis, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1428048034
APA Style (7th edition)
Gandharva, Kumar.
Study of Effect of Coverage and Purity on Quality of Learned Rules.
2015. University of Cincinnati, Master's thesis.
OhioLINK Electronic Theses and Dissertations Center
, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1428048034.
MLA Style (8th edition)
Gandharva, Kumar. "Study of Effect of Coverage and Purity on Quality of Learned Rules." Master's thesis, University of Cincinnati, 2015. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1428048034
Chicago Manual of Style (17th edition)
Abstract Footer
Document number:
ucin1428048034
Download Count:
389
Copyright Info
© 2015, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.