Mining approximate dependencies using partitions on similarity-relation-based fuzzy databases

We present a data mining technique for determining approximate dependencies in similarity-relation-based fuzzy databases. The similarity relation-based fuzzy data model is most suitable for describing analogical data over discrete domains, in addition to fuzzy set-based fuzzy data models. Approximate dependency is an extension of functional dependency such that equality of tuples is extended and replaced with the notion of equivalence class. The approximate dependency we define can capture more real-world integrity constraints than most existing functional dependencies on fuzzy databases. A level-wise mining technique is adopted for the search of all possible nontrivial minimal approximate dependencies based on equivalence classes of attribute values. An algorithm based on Huhtala (1998) is presented whereas other approximate types of functional dependencies introduce only conceptual viewpoints.