Fuzzy Apriori Method on Shapley Features

The Fuzzy Apriori Method (FAM) is a typical data mining tool with strong uncertainty-handling capability. However, the assumption of equal contribution of each feature is hampering its running speed and leverage of post-hoc interpretability provided by Shapley values. In this study, we improve the F AM by ranking features through their Shapley values. Each feature's contribution can be used to explain the features and the prior interpretability of feature selection. We term such a FAM as Shapley-FAM (S-FAM). The concept behind S-FAM is to compute the Shapley value for each feature before determining the fuzzy membership in the FAM. Subsequently, it identifies features with higher contributions, generates frequent itemsets and association rules and ultimately applies these rules to classify the 10 selected UCI benchmark datasets. With its simple principle, S-FAM keeps at least comparable to and even better than F AM in the sense of testing classification accuracy, rule count, and computational complexity on most of all the adopted datasets.