Author(s)

E. Titova, D. Sitnikov, O. Ryabov, B. D’Cruz

Abstract

The existing algorithms for discovering association rules in databases have been developed mainly for binary features (i.e. a feature is either present in a transaction or not). Nevertheless, in real databases it often happens that a feature can take on values from an arbitrary finite set. Such features are considered to be categorical values. In many cases continuous data is binned to obtain a set of discrete values. We suggest an algorithm for discovering generalised association rules in databases containing categorical data. This algorithm is based on building a tree of covers which allows us to generate associations with necessary support, and also to take into account updated records. To discover generalised associations we use a method of merging different branches of the cover tree for given values of a feature. The resulting rules look as follows: if the feature X takes on values from the set {a1,a2,...ai}, then feature Y takes on values from the set {b1,...,bk}. This is beneficial for small discrete feature domains. Keywords: data mining, generalized association rules, finite predicate. 1 Introduction Discovering association rules is a typical task for Data Mining. The process of generating associations consists of obtaining rules in the implicative form. Originally this task appeared in the marketing area (so called \“market basket

Keywords

data mining, generalized association rules, finite predicate.