Datascape Survey Using the Cascade Model

Abstract

Association rules have the potential to express all kinds of valuable information, but a user often does not know what to do when he or she encounters numerous, unorganized rules. This paper introduces a new concept, the datascape survey. This provides an overview of data, and a way to go into details when necessary. We cannot invoke active user reactions to mining results, unless a user can view the datascape. The aim of this paper is to develop a set of rules that guides the datascape survey. The cascade model was developed from association rule mining, and it has several advantages that allow it to lay the foundation for a better expression of rules. That is, a rule denotes local correlations explicitly, and the strength of a rule is given by the numerical value of the BSS (between-groups sum of squares). This paper gives a brief overview of the cascade model, and proposes a new method of organizing rules. The method arranges rules into principal rules and associated relatives, using the relevance among supporting instances of the rules. Application to a real medical dataset is also discussed.