A New Classification Approach Using Fuzzy Association Rules

Association rule induction is one of the most well-known approaches in data mining techniques. In this paper, a classification approach, called Association Based Classification (ABC), is proposed in which the fuzzy association rules are used for building classifiers. In this approach the fuzzy association rulesets are exploited as descriptive models of different classes. Furthermore, we introduce some measures to assess how well any new sample is matched to a ruleset. Using these measures, we declare the class corresponding to the best matched ruleset, as the label of the sample. To handle missing data, a new approach is also proposed for this algorithm. Several experiments have been carried out on standard datasets and the results are compared with some well-known classification algorithms such as KNN and Decision Tree classifiers. The behavior of the proposed algorithm in the case of incomplete data (data having missing values) has also been inspected. In most cases ABC out performs other classifiers and shows a satisfying behavior versus missing data.

Keywords:

Association Rule , Classification , Clustering , Fuzzy Set , Itemset , Matching Measure