Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


19467.pdf (2.29 MB)

ETD Abstract Container

Abstract Header

Rule Generation for Datasets with Ordinal Class Attributes

Gopal, Deepthi
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1448037390

Abstract Details

2015, MS, University of Cincinnati, Engineering and Applied Science: Computer Science.
For decades now rule based algorithms have grown to be a highly effective classification technique. They are applied to a wide range of fields such as medicine, financial data analysis and business marketing. These rule-learning algorithms seek to learn rules that can precisely predict the class labels for unseen instances. Every rule learning algorithm is different from the others in various different ways. Most of these algorithms work by trying to cover all the instances in the data set and hence generate a large-sized set of rules. Almost all the existing algorithms generate rules such that given a set of antecedent variable-value pairs, predict one class label specified as the consequent of each rule. Existing techniques do not allow the same set antecedent values to predict multiple classes in the consequent. However, if the class attribute is of ordinal type and it makes semantic sense to combine contiguous class labels as consequents of rules, then it may be possible to induce more efficient and accurate sets of rules from datasets. We have presented in this thesis an algorithm to induce such rules and have demonstrated that they are better at making predictions Most rule induction algorithms aim to cover all the instances of training data and thus are unable to control the number and quality of the generated rules. In our proposed approach we seek to generate very high purity rules by restricting the number of rules generated. This also causes some training instances to remain uncovered. When our rules fail to classify an instance because it was not covered during training we simple do not announce the class label for these situations. We study the effect of varying the rule quality and extent of training instance coverage to study the behavior of our algorithm. A detailed analysis of results on several datasets confirms that the proposed rule-learning method performs better at classification than some of the well-known existing algorithms.
Raj Bhatnagar, Ph.D. (Committee Chair)
Nan Niu, Ph.D. (Committee Member)
Carla Purdy, Ph.D. (Committee Member)
76 p.
Computer Science
apriori algorithm;

Recommended Citations

Citations

  • Gopal, D. (2015). Rule Generation for Datasets with Ordinal Class Attributes [Master's thesis, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1448037390

    APA Style (7th edition)

  • Gopal, Deepthi. Rule Generation for Datasets with Ordinal Class Attributes. 2015. University of Cincinnati, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1448037390.

    MLA Style (8th edition)

  • Gopal, Deepthi. "Rule Generation for Datasets with Ordinal Class Attributes." Master's thesis, University of Cincinnati, 2015. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1448037390

    Chicago Manual of Style (17th edition)

ucin1448037390
859
© 2015, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.