Fuzzy logic in association rule mining: limited effectiveness analysis.

This paper presents a comparative effectiveness analysis of fuzzy and non-fuzzy association rule mining (ARM). The corresponding motivation is the lack of relevant papers devoted to the effectiveness comparison between fuzzy and non-fuzzy ARM. The current research applies the results of fuzzy/non-fuzzy ARM to associative classification and uses the classification accuracy of the corresponding classifiers as the effectiveness measure. The research demonstrates that basic effectiveness comparison between fuzzy and non-fuzzy ARM does not necessarily speak in favour of fuzzy ARM. However, then the research demonstrates that fuzzy ARM has a distinctive ability to handle data inconsistencies, which results in the ability of corresponding fuzzy classifiers to not only provide class predictions but also indicate their certainty in the provided output. The research reveals some distinct correlation between classification accuracy and the degree of certainty in fuzzy associative classification: the greater certainty nearly always results in the better classification accuracy. This ability of fuzzy associative classifiers to indicate their certainty in the performed classification and the corresponding ability of fuzzy ARM to handle data inconsistencies are not applicable in non-fuzzy associative classifiers and ARM. Therefore, these abilities are used in the paper to substantiate the relevance of applying fuzzy logic in ARM. [ABSTRACT FROM AUTHOR]