Zbigniew W. Ra´s1,2 and Agnieszka Dardzi´nska3 1 Univ. of North Carolina, Dept. of Computer Science, Charlotte, NC, 28223, USA 2 Polish-Japanese Institute of Information Technology, 02-008 Warsaw, Poland 3 Bialystok Technical Univ., Dept. of Computer Science, 15-351 Bialystok, Poland An action rule is a rule which can be extracted from a decision system S and it describesa possible transition of objects from one decision class to another. Formally, it is definedas a term [(ω) (α → β)] (φ → ψ), where ω is the header of the rule which isa conjunction of fixed classification feature, called stable features, (α → β) representsproposed changes in values of flexible features, and (φ → ψ) is a desired effect of theaction. The discovered knowledge provides an insight of how values of some attributesneed to be changed in S so the undesirable objects can be shifted to a desirable group.
The notion of a cost of action rule will be introduced.
By meta-actions associated with S we mean higher level concepts representing ac- tions. Meta-actions, when executed, are expected to trigger changes in values of someattributes in S. In medical domain, taking a drug is a classical example of a meta-action.
For instance, Lamivudine is used for treatment of chronic hepatitis B. It improves theseroconversion of e-antigen positive hepatitis B and also improves histology staging ofthe liver but at the same time it can cause a number of other symptoms. This is whydoctors order certain lab tests to check patient’s response to that drug.
The influence matrix is used to describe the relationship between meta-actions and the expected changes within classification attributes. It should be mentioned that ex-pert knowledge concerning meta-actions involves only classification attributes. Now,if some of these attributes are correlated with the decision attribute, then any changein their values will cascade to the decision attribute through this correlation. One ofthe goals of action rule discovery is to identify all correlations between classificationattributes and the decision attribute.
To reduce the number of values for numerical attributes in S we use a classical method based either on entropy or Gini index resulting in a hierarchical discretization.
Classification attributes are partitioned into stable and flexible. Before we use any flex-ible attribute in the process of a decision tree construction, all stable attributes have tobe used first. This way the decision table is split into a number of decision subtablesleading to them from the root of the tree by uniquely defined pathes built from stableattributes. Each path defines a header in all action rules extracted from the correspond-ing subtable. Initial testing shows that the action rules built that way are more compact(have larger intervals) than action rules built with prior discretization of the decisiontable done for instance by Rough Sets Exploration System.


Abstract was misleading [Letter]. BMJ 2006;332:1272. van antidepressiva bij de behandeling van depressie. 8. Bogaert M, Maloteaux JM. Gecommentarieerd Genees-Systematisch onderzoek naar de gegevens in de liter-middelenrepertorium. BCFI, 2006. atuur. Syntheserapport. Consensusvergadering RIZIV, 9. Bupropion, pas pendant la grossesse non plus. Rev Prescr Brussel 2006 (op te vr


DES SCIENCES ET DES NOUVELLES TECHNOLOGIES Adopté le 16 mars 2005 Texte original en anglais **************************************************************************************************************************** ASPECTS ÉTHIQUES DES IMPLANTS TIC DANS LE CORPS HUMAIN avis élaboré à l’initiative directe du GEE Rapporteurs: Prof. Stefano Rodotà et Prof. Rafael Capur

Copyright © 2010-2018 Pharmacy Drugs Pdf