Simulated Assessment of Ripple Round Rules
A new Ripple Down Rules based methodology which allows for the creation of rules that use classifications as conditions has been developed, and is entitled Multiple Classification Ripple Round Rules (MCRRR). Since it is difficult to recruit human experts in domains which are appropriate for testing this kind of method, simulated evaluation has been employed. This paper presents a simulated evaluation approach for assessing two separate aspects of the MCRRR method, which have been identified as potential areas of weakness. Namely, “Is the method useful in practice?” and “Is the method acceptable, computationally?” It was found that the method appears to be of value in some, but not many, “traditional” multi-class domains, and that due to computational concerns with one aspect of the method it is considered unsuitable for domains with a very large number of cases or rules. These issues are discussed and solutions are proposed.
Keywordsripple down rules multiple classification round configuration knowledge acquisition simulated expert assessment
Unable to display preview. Download preview PDF.