Date of Award

12-2021

Document Type

Thesis

Degree Name

M.S.

Department

Computer and Information Science

Committee Chair

Ryan Benton, Ph.D.

Abstract

Action rules describe a system’s required transitions to achieve a desired class transition. Composed of stable attributes (such as age) and flexible attributes (such as interest rate), these transition observations are particularly interesting because they can give insight on a change in the system such as going from an unfavorable state to a favorable one. In the domain of action rules, what is considered rare has not been formally defined. To form a definition for rare action rules, we investigate the way that rarity has been defined in a similar domain: association rule mining. Association rule mining is a pattern mining technique that generates association rules which describe object relationships, or correlations, with one another. Although rarity has numerous definitions in this realm, they do not translate directly into action rules. This research proposes a definition for rarity in action rules and two algorithms: a consequent constraint algorithm to generate rare action rules and an attribute analysis algorithm. The rule generation algorithm utilizes a user provided consequent paired with support to generate the rare action rules and the attribute analysis algorithm uses confidence to identify potentially interesting attribute transitions with respect to a class transition.

Share

COinS