Formal Frame for Data Mining with Association Rules – a Tool for Workflow Planning

The goal of this extended abstract is to contribute to the forum for research on construction of data mining workflows. We briefly introduce a formal framework called FOFRADAR (FOrmal FRAmework for Data mining with Association Rules) and then we outline how it can be used to control a workflow of data mining with association rules. We consider this relevant to associative classifiers that use association rule mining in the training phase [3]. We deal with association rules φ ≈ ψ where φ and ψ are general Boolean attributes derived from columns of analyzed data matrices. Symbol ≈ is called 4ft-quantifier and it stands for a condition concerning a contingency table of φ and ψ [6]. Such rules are more general than rules introduced in [1]. We consider data mining process as described by the well known CRISP-DM methodology. The FOFRADAR is introduced in [5]. Its goal is to formally describe a data mining process such that domain knowledge can be used both in formulation of reasonable analytical questions and in interpretation of resulting set of association rules. No similar approach to dealing with domain knowledge in data mining is known to the authors. An application of the FOFRADAR in data mining workflows is outlined here for the first time.