Change search
ReferencesLink to record
Permanent link

Direct link
On Handling Conflicts between Rules with Numerical Features
Stockholm University, Faculty of Social Sciences, Department of Computer and Systems Sciences.
2006 (English)In: Proceedings of the 21st ACM Symposium on Applied Computing. (SAC-06), Dijon: Association for Computing Machinery (ACM), 2006, 37-41 p.Chapter in book (Other academic)
Abstract [en]

Rule conflicts can arise in machine learning systems that utilise unordered rule sets. A rule conflict is when two or more rules cover the same example but differ in their majority classes. This conflict must be solved before a classification can be made. The standard methods for solving this type of problem are to use naive Bayes to solve the conflict or using the most frequent class (CN2). This paper studies the problem of rule conflicts in the area of numerical features. A novel family of methods, called distance based methods, for solving rule conflicts in continuous domains is presented. An empirical evaluation between a distance based method, CN2 and naive Bayes is made. It is shown that the distance based method significantly outperforms both naive Bayes and CN2.

Place, publisher, year, edition, pages
Dijon: Association for Computing Machinery (ACM), 2006. 37-41 p.
Keyword [en]
Rule Learning, Rule conflicts, Numerical features
National Category
Natural Sciences
URN: urn:nbn:se:su:diva-25760DOI: 10.1145/1141277.1141284ISBN: 1-59593-108-2OAI: diva2:200464
Available from: 2006-02-16 Created: 2006-02-16 Last updated: 2012-03-29Bibliographically approved
In thesis
1. Methods of solving conflicts among induced rules
Open this publication in new window or tab >>Methods of solving conflicts among induced rules
2006 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

When applying an unordered set of classification rules to classify an example, there may be several applicable rules with conflicting conclusions regarding the most probable class to which the example belongs. This problem of having rules assigning different classes to the same example must be addressed, if a classification is to be made. The standard methods of resolving such conflicts include using the most frequent class in the examples covered by the conflicting rules and using naive Bayes to calculate the most probable class.

This thesis presents four papers, in each of which the problem of conflicting rules is addressed. In the first paper, a method that bridges the gap between Bayes rule and naive Bayes is presented. The second paper presents a data driven method for resolving rule conflicts, and the third paper explores this data driven approach further. In the last paper, a method for resolving rule conflicts in domains where the examples have numerical features is presented.

For all new methods of solving rule conflicts, it is shown that the novel methods outperform the standard methods. A correlation between the novel methods performance and their computational cost is found: usually the more costly methods obtain a higher accuracy than the less costly methods.

Place, publisher, year, edition, pages
Kista: Institutionen för data- och systemvetenskap (tills m KTH), 2006. 93 p.
Report Series / Department of Computer & Systems Sciences, ISSN 1101-8526 ; 06-005
machine learning, rule conflicts
National Category
Information Science
urn:nbn:se:su:diva-855 (URN)91-7155-209-X (ISBN)
Public defence
2006-03-10, sal A, Forum, Isafjordsgatan 39, Kista, 13:00
Available from: 2006-02-16 Created: 2006-02-16Bibliographically approved

Open Access in DiVA

No full text

Other links

Publisher's full text
By organisation
Department of Computer and Systems Sciences
Natural Sciences

Search outside of DiVA

GoogleGoogle Scholar
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Altmetric score

Total: 31 hits
ReferencesLink to record
Permanent link

Direct link