Selection of decision rules based on attribute ranking

Article type: Research Article

Authors: Stańczyk, Urszula^*

Affiliations: Institute of Informatics, Silesian University of Technology, Gliwice, Poland

Correspondence: [*] Correspondence to: Urszula Stańczyk, Institute of Informatics, Silesian University of Technology, Akademicka 16, 44-100 Gliwice, Poland. Tel.: +48 32 237 21 51; Fax: +48 32 237 27 33; Email: [email protected].

Abstract: Rule classifiers enhance understanding of data and enable to represent learned knowledge in an explicit structural form. To further improve this understanding and generalisation properties of predictors, the process of selection of decision rules can be executed. The paper presents a post-processing approach to this task. In a proposed research framework there were exploited rankings of attributes the rules refer to, thus transferring the established importance of features to the rules based on them. The attributes were weighted by selected statistical measures and machine learning algorithms. The rule classifiers were constructed in Dominance-Based Rough Set Approach and employed in the domain of stylometry for the task of authorship attribution.

Keywords: Rule classifier, selection of rules, attribute ranking, stylometry, authorship attribution, DRSA

DOI: 10.3233/IFS-151620

Journal: Journal of Intelligent & Fuzzy Systems, vol. 29, no. 2, pp. 899-915, 2015