Information Technology Reference
In-Depth Information
algorithm was more ecient then the action forest algorithm. The generated
E-action rules by both algorithms are the same. The confidence of E-action
rules is higher than the confidence action rules.
9.4
Conclusion
E-action rules are structures that represent actionability in an objective way. The
strategy used to generate them is data driven and domain independent because
it does not depend on domain knowledge. Although the definition of E-action
rules is purely objective, we still can not get rid of some degree of subjectivity
in determining how actions can be implemented. To build E-action rules, we
divide all attributes into two subsets, stable and flexible. Obviously, this partition
has to be done by users who decide which attributes are stable and which are
flexible. This is a purely subjective decision. A stable attribute has no influence
on change, but any flexible attribute may influence changes. Users have to be
careful judging which attributes are stable and which are flexible. If we apply E-
action rule on objects then it shows how values of their flexible features should be
changed in order to achieve their desired re-classification. Stable features always
will remain the same. Basically, any E-action rule identifies a class of objects
that can be reclassified from an undesired state to a desired state by properly
changing some of the values of their flexible features. How to implement these
changes often depends on the user. If the attribute is an interest rate on the
banking account then banks can take appropriate action as the rule states (i.e.,
change lower interest rate to 4.75%). In this case, it is a purely objective action.
However, if the attribute is a fever then doctors may lower the temperature by
following a number of different actions. So, this is a purely subjective concept.
Basically, we cannot eliminate some amount of subjectivity in the process of
E-action rules construction and implementation.
Acknowledgement
This research was partially supported by the National Science Foundation under
grant IIS-0414815.
References
1. Adomavicius, G., Tuzhilin, A.: Discovery of actionable patterns in databases: the
action hierarchy approach. In: Proceedings of KDD 1997 Conference, Newport
Beach, CA. AAAI Press, Menlo Park (1997)
2. Chmielewski, M.R., Grzymala-Busse, J.W., Peterson, N.W., Than, S.: The rule
induction system LERS - a version for personal computers. In: Foundations of
Computing and Decision Sciences. vol. 18(3-4), Institute of Computing Science,
Technical University of Poznan, Poland, pp. 181-212 (1993)
3. Geffner, H., Wainer, J.: Modeling action, knowledge and control. In: Prade, H.
(ed.) ECAI 1998, Proceedings of the 13th European Conference on AI, pp. 532-
536. John Wiley & Sons, Chichester (1998)
Search WWH ::




Custom Search